Master thesis, 15 ECTS
Master’s Programme in Economics / Master thesis II, 15 ECTS
Economic Policy
Uncertainty and the
Swedish stock market
A quantitative study on how Economic Policy
Uncertainty affects the Swedish stock market
Acknowledgements
Writing this thesis has been very interesting and educational. I would like to express my profound gratitude to my thesis supervisor Thomas Aronsson for providing me with guidance and feedback. I would also like to thank my family and friends for all the unfailing support during the last couple of months.
With that said, I wish you a continued pleasant read.
Abstract
This study examines the effect of Economic Policy Uncertainty (EPU) on the Swedish stock market, both in the short-run and the long-run. To do this, the Economic Policy Uncertainty index, developed by Armelius et al. (2016), is used. This index was constructed with the purpose of measuring uncertainty related to economic policy. EPU may make firms postpone important investments or decisions which could affect financial outcomes and cash flow in the future. As future cash flows are often correlated with the stock prices, this may put downward pressure on the price (Gulen & Ion, 2014). EPU may also affect the factors determining discount rates, such as interest rates, inflation, and risk premiums (Pastor and Veronesi, 2013).
The results in this study may help investors and large institutions to understand how stocks are priced during times of economic policy uncertainty. Also, if the effect of the EPU is significantly negative, it will give authorities and the government an incentive to maintain transparency to make the markets less volatile.
Earlier studies suggest that EPU has a negative effect on stock market returns in several countries such as Australia, Canada, China, Japan, Korea, and the US (Arouri et al., 2016 & Christou et al., 2017). The findings in this study suggest that EPU has a negative effect on the Swedish stock market in the short run. The effect is greatest during the same month the uncertainty shock occurs. During the first and second month after, the effect is still negative, but smaller. However, when analysing possible cointegration among the variables in the model them, no long-run relationship can be found between EPU and the stock market index.
Table of content
1. Introduction ... 1
2. Theoretical background ... 3
2.1 Stock prices and returns ... 3
2.2 Effect of EPU on the stock market. ... 4
2.3 Summary ... 6
3. Methodology ... 7
3.1 Data ... 7
Figure 1. Economic Policy Uncertainty and the Swedish Stock Market ... 8
Table 1. Descriptive statistics ... 9
3.2 Empirical model ... 10
3.3 Time series analysis ... 10
3.3.1 Autoregressive series ... 10
3.3.2 Stationary and non-stationary series ... 11
3.3.3 Optimal lag order selection ... 13
3.3.4 Cointegration ... 13
3.4 Model specification ... 15
3.4.1 Bounds Test ... 17
3.4.2 Diagnostic tests ... 17
4. Results ... 19
Table 2. Bounds Cointegration Test – ARDL(2, 3, 2, 0, 0) ... 19
Table 3. ARDL model with Error Correction Parametrization ... 20
5. Discussion and concluding remarks ... 22
5.1 Limitations ... 22
5.2 Conclusions ... 23
References ... 24
1. Introduction
This study examines the effect of Economic Policy Uncertainty (EPU) on the Swedish stock market. It is well known that the stock market is very unpredictable in the short run. Stock prices are basically based on expectations about the future. In the long run, the stock price is expected to reflect the company’s fundamentals, such as profitability, liabilities, and growth potentials. If the company is constantly growing and increasing its profits and dividends, the stock price is expected to follow. These expectations are in turn based on all the information currently available for the market. In the short run, however, the stock price is far more unpredictable and can increase or decrease without any real change in the fundamentals of the company (Cornell, 1999). There is a range of different factors and information that can affect the stock market in the short run. If, for example, some positive news about a company or the market comes out, one could expect the stock price to rise. The information provided is however far from perfect and can be difficult to interpret. There are always uncertainties about the future, and the economic actors are likely to respond to these uncertainties. In turn, these responses affect prices.
The main purpose of this study is to examine and determine the effect of economic policy uncertainty on the Swedish stock market, both in the long-run and short-run. To do this, the Economic Policy Uncertainty index, developed by Armelius et al. (2016), is used. This index was constructed with the purpose of measuring uncertainty related to economic policy. To construct the index the authors followed the methodology developed by Baker et al. (2016). The results will provide an understanding on whether the effect is negative, positive, or ambiguous. This information may help investors and large institutions to understand how stocks are priced during times of economic policy uncertainty. Also, if the effect of the EPU is significantly negative, it will give authorities and the government an incentive to maintain transparency to make the stock market less volatile.
this may put downward pressure on the price (Gulen & Ion, 2013). EPU may also affect the factors determining discount rates, such as interest rates, inflation, and risk premiums (Pastor and Veronesi, 2013). Arouri et al. (2016) found that EPU has a negative effect on the US stock market returns throughout the period 1900-2014. Similar results were found by Antonakakis et al. (2013). Their findings suggest that the dynamic correlation between policy uncertainty and the S&P 500 returns is consistently negative over time, except during the great financial crisis. Christou et al. (2017), found that EPU has a negative effect on the stock market returns in Australia, Canada, China, Japan, Korea, and the US. All studies that are referred to use the EPU index either constructed by Baker et al. (2016), or constructed using the same method as Baker et al. It is the same type of index that is used in this study, which will contribute to this growing literature by examining the effect of EPU on the Swedish stock market.
The findings in this study suggest that EPU has a negative effect on the Swedish stock market index in the short run. The effect is greatest during the same month the uncertainty shock occurs. During the first and second month after, the effect is still negative, but smaller. However, when analysing a possible cointegration between EPU and the stock market index, no long-run relationship can be established. These results suggest that EPU is not part of the fundamental characteristic that is expected to reflect stock prices in the long run.
2. Theoretical background
2.1 Stock prices and returns
When it comes to valuing stocks, there are some different models one can use. Although some of the methods look different, the main purpose of most of them is to measure the present value of future incomes for the stockholder. One commonly used model to value stocks is the Discounted Dividend Model, also known as the Gordon Growth Model (Gordon, 1956). The main assumption of the model is that a stock is worth the sum of the present value of all future dividends. A stock price will, therefore, be affected by both the size of the dividends and the discount rate (equation 1). As the discount rate is in the denominator, a larger discount rate would decrease the value of the stock. Note that this method can only be used if the company is paying out dividends. Another popular method used for stock valuation is to calculate the present value of future cash flows (Danthine & Donaldson, 2001). It is done by estimating all future cash flows and then discount them back to their present values (equation 2). This value can then be divided by the number of stocks outstanding to calculate the intrinsic value of the stock. As in the Gordon Growth Model, a discount rate is used to calculate the cash flows present value. 𝑃𝑟𝑖𝑐𝑒0 = ∑ 𝐷𝑖𝑣𝑖𝑑𝑒𝑛𝑑0(1 + E[𝐷𝑖𝑣𝑖𝑑𝑒𝑛𝑑 𝑔𝑟𝑜𝑤𝑡ℎ]) 𝑡 (1 + 𝑑𝑖𝑠𝑐𝑜𝑢𝑛𝑡 𝑟𝑎𝑡𝑒)𝑡 ∞ 𝑡=1 (1)
In equation (1), 𝐷𝑖𝑣𝑖𝑑𝑒𝑛𝑑0 denotes the dividend at time 0 and t denotes time.
𝑉𝑎𝑙𝑢𝑒0 = ∑ E[𝐶𝑎𝑠ℎ 𝑓𝑙𝑜𝑤𝑡] (1 + 𝑑𝑖𝑠𝑐𝑜𝑢𝑛𝑡 𝑟𝑎𝑡𝑒)𝑡 ∞
𝑡=1
(2)
discount rate that corresponds to the expected rate of return for the stock of interest. The discount rate is a function of the risk-free rate and a so-called risk premium (see equation 3). The risk premium represents a compensation for investors in return for holding a risky asset. Another difficulty with the models is to estimate future dividends and cashflows.
𝐷𝑖𝑠𝑐𝑜𝑢𝑛𝑡 𝑟𝑎𝑡𝑒 = 𝑟𝑖𝑠𝑘𝑓𝑟𝑒𝑒 𝑟𝑎𝑡𝑒 + 𝜋 (3)
where, 𝜋 is the risk premium.
By looking at equation (3), one can conclude that if the risk premium would increase the discount rate would also increase.
According to earlier empirical studies, the risk premium may be modelled as a linear combination of different factors, such as inflation and GDP growth (Danthine & Donaldson, 2001). Chen et al. (1986) also suggest industrial production and inflation expectations. These factors will be discussed more thoroughly in section 3.1.
2.2 Effect of EPU on the stock market.
Some empirical studies have been carried out to examine the relationship between EPU and the stock market index. Uncertainty may affect stocks through many different channels. Firms may postpone important investments or decisions which could affect financial outcomes and cash flows in the future. As future cash flows are often correlated with the stock price today, this may put downward pressure on the price (Gulen & Ion, 2013). Recall that stock prices are theoretically a function of various factors. One factor is the discount rate, which can be modelled as a function of interest rates and a risk premium. Inflation should also be highlighted, as it, in theory, is affecting interest rates (Carlin & Soskice, 2006). Pastor and Veronesi (2013), found that economic uncertainty significantly affects all these factors. Given these results, one would expect that a change in EPU has an impact on stock prices and returns.
Arouri et al. (2016), confirmed this in an empirical study. Using an Autoregressive distributed lag (ARDL) model, they quantified the effects of a change in EPU on the S&P 5001 returns
1 Standard & Poor’s 500: measures the performance of 500 large companies listed on the Stock Exchanges in the
over the period 1900 – 2014. Their findings suggest that an increase in policy uncertainty reduces stock returns in the short run. They also found that the effect is larger during periods of high volatility in the stock market. Similar results were found by Antonakakis et al. (2013). Using a Dynamic Conditional Correlation (DDC) model, and data between January 1987 – January 2013, they found that the dynamic correlation between policy uncertainty and the S&P 500 returns is consistently negative over time, except during the great financial crisis. There are also some studies examining the effect of EPU on stock market returns in other countries. Christou et al. (2017), found that EPU has a negative effect on the stock market returns in the short run in Australia, Canada, China, Japan, Korea, and the US. They used a Panel Vector Autoregression (VAR) model and data between January 1998 – December 2014. A contradictory result compared to those mentioned above was found by Chang et al. (2015). They examined the relationship between policy uncertainty and stock market returns in Canada, France, Germany, Italy, Spain, the UK, and the US. Using a Bootstrap Panel Causality Test, they found that uncertainty has an overall positive effect on the stock market in the US and UK. The sign of the effect is however based on the sum-of-coefficients on all the lags and does not necessarily imply that the effect is positive on impact. Kang and Ratti (2013), for example, found that the effect of EPU on stock market returns is negative at impact, but becomes positive at longer lags. All studies that are referred to use the EPU index either constructed by Baker et al. (2016), or constructed using the same methodology as Baker et al. Moreover, most of them only focus on the short-run dynamics and do not test whether cointegration exists among EPU and the stock market index.
2.3 Summary
In this section, the theoretical framework has been presented. Stock prices and returns are difficult to predict, and it may be even more difficult to explain the reasons behind short term fluctuations. The purpose of this study is to examine the effect of Economic Policy Uncertainty on the Swedish stock market. Both the long-run relationship and short-run dynamics is analysed. Given the results from the previous studies presented in section 2.2, it is expected that EPU has a negative effect on the stock market index. The expectation is based on the idea that economic policy uncertainty has a positive effect on the risk premium demanded from investors, which in turn will affect the discount rate. As explained in section 2.1, a stock price is in theory a function of a discount rate, which is used to calculate the present value of something from the future. If the discount rate increases, the present value decreases, which could put downward pressure on the stock price.
3. Methodology
3.1 Data
This study is based on monthly data from January 1995 to December 2019. The two main variables in this study are the Swedish Economic Policy Uncertainty (EPU) index and the OMX Stockholm all-share index (OMXSPI). The economic policy uncertainty index series is constructed by Armelius et al. (2016) with the purpose to measure uncertainty related to economic policy. A higher value of the index indicates more uncertainty. To construct the index the authors followed the methodology developed by Baker et al. (2016), which is the method used to create the indexes used in the previous studies discussed in section 2.2. Briefly explained, the researchers obtained monthly counts of articles containing key Swedish terms that are related to the economy (E), policy matters (P) and uncertainty (U)2. The raw data count is then scaled by the number of articles that contained keywords related to the economy component for the same newspaper and month. These scaled counts are then averaged across the four newspapers by month to obtain an index. The final index contains articles from Aftonbladet, Expressen, Dagens Industri, and Svenska Dagbladet. OMXSPI is obtained from the Nasdaq OMX Nordic database and contains all stocks traded on the Stockholm Stock Exchange. The time series data contains monthly averages of the index and ranges from January 1995 to December 2019. Figure 1 displays the evolution of the EPU index and the Swedish stock market.
2 The terms that are counted are ekonomi or ekonomisk (Economic), riksbank, centralbank, regering, departement
Figure 1. Economic Policy Uncertainty and the Swedish Stock Market
inventory. GDP growth rate is excluded in the study since it does not come in monthly form. PMI will, however, capture some of the effects from GDP growth, as both measures economic activity. The index also captures the level of employment, which has been shown to have a significant effect on the stock market (Arouri et al., 2016). Worth noting is that the index is a qualitative index based on monthly surveys filled out by managers in the private sector. The choice of control variables is based on economic theory as well as previous studies. Recall, from section 2.1, that a stock price depends on the discount rate that is used to calculate the present value of future dividends and cash flows. This discount rate is assumed to depend on the expected rate of return, which in turn is a function of the risk-free rate and a risk premium. STIBOR3M is used as a proxy for this risk-free rate and will, therefore, control for fluctuations in it. Interest rates are also expected to affect the risk premium demanded by investors. Higher interest rates, for example, affect not only firms profit but also potential investments. An alternative to the nominal STIBOR rate could be the real STIBOR rate, but since nominal rates are used more frequently in earlier studies, the choice fell on the nominal STIBOR rate. The risk-free rate is also expressed in nominal terms in the theoretical models mentioned in section 2.1. Furthermore, inflation is controlled for via the CPI. Inflation is also suggested by Danthine and Donaldson (2001) as a factor that affects the risk premium. Sharpe (1999) found that higher expected inflation will increase the risk premium, and hence also the expected rate of return. Finally, PMI is used to control for economic activity at the aggregate level.
The three control variables used in this study are also used in other studies examining the effect of economic policy uncertainty on the stock market (see for example Arouri, et al. (2016) and Christou et al. (2017)). In Table 1, descriptive statistics for the variables are presented.
Table 1. Descriptive statistics
Variable Number of
Observations Mean
Standard
deviation Min Max
OMXSPI 300 321.70 145.59 84.29 669.88 EPU 300 94.35 19.15 53.73 156.73 PMI 300 54.44 5.65 34 65.1 CPI 300 290.71 25.32 251.3 337.68 STIBOR3M (%) 300 2.58 2.29 -0.6145 9.29
3.2 Empirical model
To examine the effect of EPU on the Swedish stock market in both the long-run and short-run, an Error Correction Model (ECM) derived from an ARDL model is used. To ease the interpretation of the results, all variables are transformed into their natural logarithmic form, except STIBOR3M as it is negative in some periods. This also reduces the risk of heteroscedasticity. The model is specified in equation (4). In section 3.2 and 3.3, the reasoning behind this approach will be discussed. The econometric model can be written as follows:
𝛥 𝑙𝑛(𝑂𝑀𝑋𝑃𝐼)𝑡 = 𝑐0+ ∑𝛼0𝑖∆ 𝑙𝑛(𝑂𝑀𝑋𝑃𝐼)𝑡−𝑖 𝑝−1 𝑖=1 + ∑𝛼1𝑖∆ln (𝐸𝑃𝑈)𝑡−𝑖 𝑞1−1 𝑖=0 + ∑𝛼2𝑖∆ln (𝐶𝑃𝐼)𝑡−𝑖 𝑞2−1 𝑖=0 + ∑𝛼3𝑖∆ln (𝑃𝑀𝐼)𝑡−𝑖 𝑞3−1 𝑖=0 + ∑𝛼4𝑖∆𝑆𝑇𝐼𝐵𝑂𝑅3𝑀𝑡−𝑖 𝑞4−1 𝑖=0 + 𝜆𝐸𝐶𝑇𝑡−1+ 𝜀𝑡 (4) where, 𝐸𝐶𝑇𝑡−1 = 𝑙𝑛(𝑂𝑀𝑋𝑃𝐼)𝑡−1− 𝜃1𝑙𝑛(𝐸𝑃𝑈)𝑡− 𝜃2𝑙𝑛(𝐶𝑃𝐼)𝑡− 𝜃3𝑙𝑛(𝑃𝑀𝐼)𝑡− 𝜃4𝑆𝑇𝐼𝐵𝑂𝑅3𝑀𝑡, 𝜃𝑗 is the long-run parameter, for variable j and 𝜆 the speed to adjustment coefficient. 𝛼𝑗𝑖 represent short-run dynamics for variable j, 𝜀𝑡 is an error term and (𝑝, 𝑞1, 𝑞2, 𝑞3, 𝑞4) the lag length.
3.3 Time series analysis
Time series data is a sequence of data points observed over a given time. As time series often possess specific properties, such as trends and structural breaks, a simple linear regression may not be appropriate for the analysis. It is, therefore, important to understand the behaviour of all the variables.
3.3.1 Autoregressive series
returns on the Swedish stock market index can be described as an AR-process of order 1, AR (1). This means that yesterday’s return of the index has a direct effect on today’s return, i.e.,
𝑟𝑒𝑡𝑢𝑟𝑛𝑡 = 𝛽0+ 𝛽1𝑟𝑒𝑡𝑢𝑟𝑛𝑡−1+ 𝜖𝑡 (5)
where, 𝛽0 is a constant, 𝛽1 is the coefficient for 𝑟𝑒𝑡𝑢𝑟𝑛𝑡−1, and 𝜖𝑡 is an error term.
This study examines the effect of EPU on the Swedish stock market. If the Swedish stock market follows an AR-process of any order, it is important to include this in the model. If it is excluded the coefficients for the other variables will be biased.
3.3.2 Stationary and non-stationary series
Another important concept within time series analysis is stationarity. A time series is stationary when its probability distribution does not change over time. This means that the statistical properties, such as the mean and variance, do not change over time. Time series that are stationary tend to return to their long-run average. If a time series has a so-called unit root, it is non-stationary (Stock & Watson, 2015).
Non-stationarity is very common among macroeconomic variables such as GDP, consumption, investment, etc. The problem arises when the non-stationary variables are regressed on each other. The conventional t-tests may suggest that we reject the null hypotheses of no relationship when, in fact, there is no relationship. This type of regression is called a spurious regression. The strong regression relationship may only exist due to underlying characteristics of the time series, e.g. trends (Greene, 2012). It is, therefore, important to determine whether the time series included in this study are stationary or not. Graphs can give a preliminary idea about the stationarity of the series, but unit root tests are usually used for formal testing.
∆𝑌𝑡= 𝛽0+ 𝛿𝑌𝑡−1+ ∑ 𝛽𝑖∆𝑌𝑡−𝑖+ 𝑒𝑡 𝑝
𝑖=1
(6)
where, ∆𝑌𝑡 is the first difference between 𝑌𝑡 and 𝑌𝑡−1 and 𝑝 the number of lags.
The null hypothesis of the ADF test is that 𝛿 = 0, which is tested against the alternative of 𝛿 < 0. If the null hypothesis is rejected, the series is stationary. The number of lags is determined by a trial and error procedure. First, an upper bound for p is set using a method suggested by Schwartz (1989): 𝑝𝑚𝑎𝑥 = (12 × ( 𝑇 100) 1 4 ) (7)
where, 𝑇 is the number of observations.
When 𝑝𝑚𝑎𝑥 is set, the ADF regression (equation 6) is estimated with 𝑝 = 𝑝𝑚𝑎𝑥. The idea is to test whether the coefficient for the last lag is significant. If it is, 𝑝 is set to 𝑝𝑚𝑎𝑥. If not, the regression is estimated with 𝑝 = 𝑝𝑚𝑎𝑥− 1. This is repeated until a rejection occurs (Ng and Perron, 1995).
In some cases, a time series can be non-stationary, but stationary around a deterministic trend. To test for this, the trend must be added as a regressor in the Dickey-Fuller regression:
∆𝑌𝑡 = 𝛽0+ 𝜌𝑡 + 𝛿𝑌𝑡−1∑ 𝛽𝑖∆𝑌𝑡−𝑖+ 𝑒𝑡 𝑝
𝑖=1
(8)
where, 𝑡 is the trend.
stationary by taking the first difference. Any long-run relationship/information may, however, be lost by doing so (Greene, 2012).
3.3.3 Optimal lag order selection
As in the ADF test, it is important to determine the optimal number of lags in the main model (equation 4). Too few lags are can decrease forecast accuracy, while too many can increase the estimation uncertainty. One approach to get around this dilemma is to by minimizing the value of an “information criteria”. The criteria used in this study is the AIC, Akaike information criteria, (Akaike, 1974) and is defined as follows:
𝐴𝐼𝐶 = −2𝑙𝑛[𝑀𝐿] + 2𝑘 (9)
where, 𝑀𝐿 is the maximum likelihood of the model and 𝑘 is the number of independently adjusted parameters in the model.
Lags are considered in this study because there are reasons to believe that a change in the explanatory variables could affect the dependent variable, not only when the change occurs, but also in the future. Also, as mentioned in section 3.2.1, in time series data it is common that the variables are serially correlated. Lags must, therefore, be included to control for this.
3.3.4 Cointegration
variables that are integrated of order two or higher. To describe the approach, the simple ARDL(p,q) model can be considered:
𝑌𝑡= 𝑐𝑜+ ∑ 𝛼𝑖𝑌𝑡−𝑖 𝑝 𝑖=1 + ∑ 𝛽𝑖𝑋𝑡−𝑖 𝑞 𝑖=0 + 𝜀𝑡 (10)
Through a linear transformation, the ARDL(p,q) model of equation (10) can be transformed into a dynamic ECM specified as:
∆𝑌𝑡 = 𝑐0+ ∑ 𝛿𝑖∆𝑌𝑡−𝑖 𝑝−1 𝑖=1 + ∑ 𝜔𝑖∆𝑋𝑡−𝑖 𝑞−1 𝑖=0 + 𝛾1𝑌𝑡−1+ 𝛾2𝑋𝑡+ 𝜀𝑡 (11)
where, 𝛿𝑖 and 𝜔𝑖 represents the short-run dynamics and 𝛾𝑗, 𝑗 = 1, 2, represents the long-run dynamics.
When equation (11) is estimated, the next step is to perform the Bounds Test. The test is essential an F-test analysing the coefficients that represent the long-run dynamics, i.e. 𝛾1 and 𝛾2. The null and alternative hypothesis is expressed as:
𝐻0 ∶ 𝛾𝑗 = 0 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑗: No cointegration exists
𝐻1 ∶ 𝛾𝑗 ≠ 0 𝑓𝑜𝑟 𝑠𝑜𝑚𝑒 𝑗: Cointegration exists among the variables
The Bounds Test assumes that the model includes both I(0) and I(1) variables. Thus, two levels of critical values are obtained. The lower bound is calculated on the assumption that all variables are I(0) while the upper bound is calculated on the assumption that all variables are I(1) (Pesaran et al., 2001). To reject the null hypothesis of no cointegration, the F-statistic must be larger than the upper bound critical value for a specific significance level. If the F-statistic is smaller than the lower bound critical value, the null hypothesis cannot be rejected. However, if the F-statistic fall in between the lower and upper bound, the test is inconclusive.
and can together with a speed of adjustment coefficient be used to rewrite the part describing the long-run dynamics in equation (11). The new equation is specified as:
∆𝑌𝑡 = 𝑐0+ ∑ 𝛿𝑖∆𝑌𝑡−𝑖 𝑝−1 𝑖=1 + ∑ 𝜔𝑖∆𝑋𝑡−𝑖 𝑞−1 𝑖=0 + 𝜆𝐸𝐶𝑇𝑡−1+ 𝜀𝑡 (12)
where, 𝐸𝐶𝑇𝑡−1= 𝑌𝑡−1− 𝜃𝑋𝑡−1. 𝜃 denote the long-run parameter. 𝜃 = −𝛾2
𝛾1.
The speed of adjustment coefficient, 𝜆, must have a negative sign. If not, the model will not converge to equilibrium. The size of the coefficient gives us information on how quickly the dependent variable returns to a long-term equilibrium.
3.4 Model specification
𝛥 𝑙𝑛(𝑂𝑀𝑋𝑆𝑃𝐼)𝑡 = 𝑐0+ ∑𝛼0𝑖∆ 𝑙𝑛(𝑂𝑀𝑋𝑆𝑃𝐼)𝑡−𝑖 𝑝−1 𝑖=1 + ∑𝛼1𝑖∆𝑙𝑛(𝐸𝑃𝑈)𝑡−𝑖 𝑞1−1 𝑖=0 + ∑𝛼2𝑖∆𝑙𝑛(𝐶𝑃𝐼)𝑡−𝑖 𝑞2−1 𝑖=0 + ∑𝛼3𝑖∆𝑙𝑛(𝑃𝑀𝐼)𝑡−𝑖 𝑞3−1 𝑖=0 + ∑𝛼4𝑖∆𝑆𝑇𝐼𝐵𝑂𝑅3𝑀𝑡−𝑖 𝑞4−1 𝑖=0 + 𝛽0𝑙𝑛(𝑂𝑀𝑋𝑆𝑃𝐼)𝑡−1 + 𝛽1𝑙𝑛(𝐸𝑃𝑈)𝑡+ 𝛽2𝑙𝑛(𝐶𝑃𝐼)𝑡+ 𝛽3𝑙𝑛(𝑃𝑀𝐼)𝑡 + 𝛽4𝑆𝑇𝐼𝐵𝑂𝑅3𝑀𝑡+ 𝜀𝑡 (13)
where, (𝑝, 𝑞1, 𝑞2, 𝑞3, 𝑞4) denotes the optimal lag lengths, determined by AIC (see section 3.2.3) The 𝛽𝑗 coefficients (j = 0, 1, …, 4) represents the long-run effects, while the short-run dynamics are captured by the 𝛼𝑗𝑖 coefficients.
3.4.1 Bounds Test
To test whether cointegration exists among the variables Bounds cointegration test is used. The test is described in section 3.2.4 and is conducted on the long-run coefficients in equation (13). The null and the alternative hypothesis of the test are expressed as:
𝐻0 ∶ 𝛽𝑗 = 0 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑗: No cointegration exists
𝐻1 ∶ 𝛽𝑗≠ 0 𝑓𝑜𝑟 𝑠𝑜𝑚𝑒 𝑗: Cointegration exists among the variables
The null hypothesis is rejected if the F-statistic is larger than the upper bound critical value and accepted if it falls below the lower bound critical value.
3.4.2 Diagnostic tests
When the model is estimated, it is important to perform some diagnostic tests to make sure the estimates are robust and efficient. To test for serial correlation in the residuals, Breusch-Godfrey Lagrange Multiplier test is used (Breusch-Godfrey, 1978). If the residuals are suffering from serial correlation, this means that they are following an AR process, as described in section 3.2.1, which can affect the standard errors of the coefficients in the regression. In turn, this makes the significance tests unreliable. In the simplest form of the Breusch-Godfrey test, the residuals are modelled as:
𝜀𝑡= 𝜌𝜀𝑡−1+ 𝑣𝑡 (15)
where, 𝑣𝑡 ~ 𝑁(0, 𝜎2)
The null hypothesis is that 𝜌 = 0, which means that no serial correlation exists. To test for heteroskedasticity, White’s test for heteroskedasticity is used (White, 1980). If the residuals are heteroskedastic, the estimated standard errors will not be accurate. The consequence could be that the wrong inferences are made. The null hypothesis is that the residuals are homoskedastic with a constant variance. If the model suffers from heteroskedasticity, the model will be re-estimated with robust standard errors.
4. Results
In this section, the results from the ARDL model presented in section 3.3 are presented. This includes the Bounds Test, diagnostic tests, and the final model. The Bounds Test on the long-run coefficients in equation (13) suggests that there exists cointegration among the variables. This can be seen in table 2. The F-statistic is 5.55, which is larger than the upper bound critical value at a 1% significance level. The validity of this model and the results from the Bounds Test were tested using the diagnostic tests described in section 3.4.2. The test results (see appendix) suggest that no serial correlation exists in the residuals and that the long-run and short-run coefficients are stable. However, the model suffers from heteroskedasticity, which makes the Bounds Test results unreliable. To control for heteroskedasticity, the model is re-estimated with robust standard errors. Using these estimates, an additional Bounds Test was conducted on the long-term coefficients. The F-statistic of the Bounds Test on the model with robust standard errors is 4.14, which is larger than the upper bound critical value at a 5 percent significance level. The conclusion is therefore that there exists cointegration among the variables.
Table 2. Bounds Cointegration Test – ARDL(2, 3, 2, 0, 0)3
𝑯𝟎 ∶ 𝜷𝒋 = 𝟎 𝒇𝒐𝒓 𝒂𝒍𝒍 𝒋 𝑯𝟏 ∶ 𝜷𝒋 ≠ 𝟎 𝒇𝒐𝒓 𝒔𝒐𝒎𝒆 𝒋
F-statistic: 5.55 F-statistic (Robust SE): 4.14 Critical values Significance level 1% I(0) I(1) 2.25% I(0) I(1) 5% I(0) I(1) 10% I(0) I(1) Critical value 3.74 5.06 3.25 4.49 2.86 4.01 2.45 3.52
Having established cointegration, the next step is to analyse the long-run relationship and the short-run dynamics using the models presented in section 3.4. The model and its estimated coefficients are presented in table 3 below.
Table 3. ARDL model with Error Correction Parametrization 𝐷𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒: ∆𝑙𝑛 (𝑂𝑀𝑋𝑃𝐼)𝑡 𝐿𝑎𝑔 𝑙𝑒𝑛𝑔𝑡ℎ: (2, 3, 2, 0, 0) 𝐴𝑑𝑗 𝑅2 = 26.52% Variable Coefficient ADJ 𝑙𝑛 (𝑂𝑀𝑋𝑆𝑃𝐼)𝑡−1 -0.042*** (0.012) Long run 𝑙𝑛 (𝐸𝑃𝑈)𝑡 𝑙𝑛 (𝐶𝑃𝐼)𝑡 𝑙𝑛 (𝑃𝑀𝐼)𝑡 𝑆𝑇𝐼𝐵𝑂𝑅3𝑀𝑡4 0.466 (0.390) 0.567 (1.606) 1.884** (0.809) -0.171*** (0.056) Short run ∆𝑙𝑛 (𝑂𝑀𝑋𝑃𝐼)𝑡−1 ∆𝑙𝑛 (𝐸𝑃𝑈)𝑡 ∆𝑙𝑛 (𝐸𝑃𝑈)𝑡−1 ∆𝑙𝑛 (𝐸𝑃𝑈)𝑡−2 ∆𝑙𝑛 (𝐶𝑃𝐼)𝑡 ∆𝑙𝑛 (𝐶𝑃𝐼)𝑡−1 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 0.179*** (0.056) -0.115*** (0.189) -0.075*** (0.019) -0.044** (-0.017) -1.326** (0.624) -1.432*** (0.632) -0.274 (0.364)
Notes: *, **, ***, significant at a 10%, 5% and 1% significance level, respectively. Robust standard errors are presented in the parenthesis.
The dependent variable in the model is ∆𝑙𝑛 (𝑂𝑀𝑋𝑆𝑃𝐼)𝑡. The estimated coefficients of the model are presented in three sections in the table. First, the speed of adjustment coefficient, 𝜆, is displayed in the ADJ section. The coefficient is negative and significant at a 1 percent significant level (-0.042). The parameter must be negative for the model to converge to the long-run equilibrium. As explained in section 3.2.4, the speed of adjustment coefficient shows how quickly the dependent variable returns to a long-term equilibrium after a shock. A high adjustment coefficient (in absolute term), indicates a faster adjustment process. In this case, the coefficient is -0.042, which means that 4.2 percent of the disequilibrium of the previous month shock is corrected back to the long-run equilibrium in the current month.
In the Long run section, the long-run parameter for each explanatory variable is displayed. The results suggest that EPU does not have a long-run effect on the Swedish stock market index. The same goes for CPI. PMI has a positive long-run effect on the stock market at a 5 percent
4 The model was also estimated using the real STIBOR3M rate. The results suggested no significant effect of the
significance level while STIBOR3M has a negative long-run effect at a 1 percent significance level.
5. Discussion and concluding remarks
In this study, the effect of economic policy uncertainty on the Swedish stock market has been empirically examined and presented. The results show that EPU has a significantly negative effect on the stock market returns in the short run. When analysing the long-run relationship, however, no cointegration could be established between EPU and the stock market. These results suggest that EPU is not part of the fundamental characteristic that is expected to reflect stock prices in the long run. The significant short-run effect is in line with both the expectations based on theory mentioned in section 2.3 and the results from the previous studies presented in section 2.2 (see for example Arouri et al., 2016 and Christou et al., 2017). Also, the effects of the control variables are in line with what theory suggests and previous studies. PMI has a positive effect in the long run on the stock market returns, which shows that a booming economy drives up stock prices. Similar results were found by Arouri, et al. (2016). CPI has a negative effect on returns in the short run, which was also found by Sharpe (1999). More specifically, Sharpe found that inflation has a positive effect on the risk premium demanded from investors, which in turn increases the discount rate. STIBOR3M has a negative effect on returns in the long run, which indicates that higher interest rates tend to decrease the value of the stocks. Although analysing the effect of the control variables are not the main purpose of the study, it is still important to check whether they are in line with what the theory says and results from previous studies.
5.1 Limitations
The index used to capture uncertainty related to economic policy is not perfect either. It is constructed by counting the number of news articles containing specific words related to uncertainty, economic, and policy. It does however not capture the magnitude of the uncertainty. This means that newspapers may print a lot of articles about an uncertainty that is not that important to the economy and fewer articles about an uncertainty that could have a serious impact on the economy.
A final drawback with the model used is that the estimated parameters are based on linear specifications. The CUSUMQ test result implies that the model does not suffer from structural breaks, but as mentioned earlier, there is still a possibility that it does. However, due to time limitations, this has not been examined.
5.2 Conclusions
The conclusion that can be drawn from this study is that economic policy uncertainty has a negative effect on the Swedish stock market returns in the short run. The findings may help investors and large institutions to understand how stocks are priced during times of economic policy uncertainty. They also show the importance of authorities and the government to maintain transparency to keep the stock market less volatile.
References
Akaike, H. (1974). A new look at the statistical model identification. Automatic Control, IEEE
Transactions on 19(6), 716-723.
Antonakakis, N., Chatziantoniou, I., & Filis, G. (2013). Dynamic co-movements of stock market returns, implied volatility and policy uncertainty. Economics Letters, 120(1), 87–92. Armelius, H., Hull, I., & Köhler, H. S. (2017). The timing of uncertainty shocks in a small open economy. Economics Letters, 155, 31-34.
Arouri, M., Estay, C., Rault, C. & Rouband, M. (2016). Economic Policy Uncertainty and Stock Markets: Long-Run evidence from the US. Finance Research Letters, May 2016.
Baker, S. R., Bloom, N., & Davis, S. J. (2016). Measuring economic policy uncertainty. The
quarterly journal of economics, 131(4), 1593-1636.
Brown, R. L., Durbin, J. & Evans, J. M. (1975). Techniques for testing the constancy of regression relationships over time. Journal of the Royal Statistical Society. Series B
(Methodological), 149-192.
Carlin, W. & Soskice, D. (2006). Macroeconomics: Imperfections, Institutions & Policies. Oxford University Press Inc.
Chang, T., Chen, W.Y., Gupta, R. & Nguyen, D. K. (2015). Are stock prices related to political uncertainty index in OECD countries? Evidence from bootstrap panel causality test. Economic
Systems 39(2), 288-300.
Chen, N.F., Roll, R., Ross, S.A. (1986). Economic forces and the stock market. J. Bus. 59, 383-404.
Christou, C., Cunado, J., Gupta, R. & Hassapis, C. (2017). Economic policy uncertainty and stock market returns in PacificRim countries: Evidence based on a Bayesian panel VAR model.
Journal of Multinational Financial Management, 40 (2017), 92-102.
Danthine, J.P. & Donaldson, J.B. (2014). Intermediate Financial Theory, 3rd edition. Academic Press, Pearson Education, Inc.
Dickey, D. A., & Fuller, W. A. (1979). Distribution of the estimators for autoregressive time series with a unit root. Journal of the American statistical association, 74(366a), 427-431. Godfrey, L. G. (1978). Testing against general autoregressive and moving average error models when the regressors include lagged dependent variables. Econometrica: Journal of the
Econometric Society, 1293-1301.
Gordon, M.J & Shapiro, E. (1956). Capital Equipment Analysis: The Required Rate of Profit.
Management Science, 3, (1) (October 1956), 102-110. Reprinted in Management of Corporate
Capital, Glencoe, Ill.: Free Press of, 1959.
Greene, W. H. (2012). Econometric analysis, 7th edition. Upper Saddle River, N.J., Prentice Hall.
Gulen, H. & Ion M. (2013). Policy Uncertainty and Corporate Investment. Working paper, Purdue University.
Hull, J. C. (2012). Options, futures, and other derivatives. Pearson Education.
Johansen, S. (1988). Statistical analysis of cointegration vectors. Journal of Economic
Dynamics and Control, Volume 12.
Johansen, S. (1991). Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models. Econometrica: Journal of the Econometric Society, 1551-1580. Kang, W., Ratti, R.A. (2013). Oil shocks, policy uncertainty and stock market returns. J. Int.
Finance. Markets Institutions Money 26, 305–318.
Naiya, I. I. & Manap, T. A. A. (2013). Structural Transformation, Poverty and Inequality in Nigeria: An ARDL Bound Testing Technique. International Journal of Economics and Finance
5(4), 141
Ng, S. Perron, P. (1995). Unit Root Tests in ARMA Models with Data-Dependent Methods for the Selection of the Truncation Lag. Journal of the American Statistical Association, Vol 90,
Issue 429 (Mars, 1995), 267-281.
Stock, J.H. & Watson, M.W. (2015). Introduction to Econometrics. Updated 3rd edition. London: Pearson Education Limited.
Schwert, G. W. (1989). Tests for Unit Roots: A Monte Carlo Investigation. Journal of Business
and Economic Statistics, 7, 147-160.
Pastor, L. & Veronesi, P. (2013). Political uncertainty and risk premia. Journal of Financial
Economics, 110, 520–545
Pesaran, M. H. & Shin, Y. (1998). An autoregressive distributed lag modelling approach to cointegration analysis. Econometric Society Monographs 31, 371-413.
Pesaran, M. H., Shin, Y., & Smith, R. J. (2001). Bounds testing approaches to the analysis of level relationships. Journal of applied econometrics, 16(3), 289-326.
White, H. (1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica: Journal of the Econometric Society, 817-838.
Data
• OMXSPI: Nasdaq OMX Nordic database. [Retrieved 2020-03-02]
• EPU: www.policyuncertainty.com [Retrieved 2020-03-02]
• PMI: www.swedbank.se [Retrieved 2020-03-25]
• CPI: Swedish Statistics database. [Retrieved 2020-03-25]
Appendix
Augmented Dickey-Fuller test – Level form
Variable Included term Test-statistic Decision OMXSPI Constant & Trend
Constant
-3.087
-1.762 Non-stationary EPU Constant & Trend
Constant
-3.126*
-2.984** Stationary
PMI Constant & Trend Constant
-5.515***
-5.529*** Stationary
CPI Constant & Trend Constant
-2.828
0.102 Non-stationary
STIBOR3M Constant & Trend Constant
-4.502***
-2.575 Non-stationary
*, **, ***, significant at a 10%, 5% and 1% significance level, respectively. Constant: Estimated accordingly to equation (6).
Constant & Trend: Estimated accordingly to equation (8).
Critical Values
1% 5% 10%
Constant
Constant & Trend
-3.43 -3.96 -2.86 -3.41 -2.57 -3.12
Reject the null hypothesis if the test-statistic is smaller than the critical value.
Augmented Dickey-Fuller test – First difference
Variable Included term Test-Statistic Decision
OMXSPI Constant -6.573*** Stationary
CPI Constant -3.741*** Stationary
STIBOR3M Constant -5.069*** Stationary
*, **, ***, significant at a 10%, 5% and 1% significance level, respectively. Constant: Estimated accordingly to equation (6).
Critical Values
1% 5% 10%
Constant -3.43 -2.86 -2.57
Breusch-Godfrey LM test for autocorrelation 𝐻0: 𝑁𝑜 𝑠𝑒𝑟𝑖𝑎𝑙 𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛
𝐻1: 𝑆𝑒𝑟𝑖𝑎𝑙 𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛
Lags Chi2 Prob > Chi2
1 0.001 0.9726
2 0.146 0.9298
3 0.348 0.9507
4 0.858 0.9306
Cumulative sum test for parameter stability 𝐻0: 𝑁𝑜 𝑠𝑡𝑟𝑢𝑐𝑡𝑢𝑟𝑎𝑙 𝑏𝑟𝑒𝑎𝑘
𝐻1: 𝑆𝑡𝑟𝑢𝑐𝑡𝑢𝑟𝑎𝑙 𝑏𝑟𝑒𝑎𝑘
Test statistic 1% 5% 10%
0.5904 1.1430 0.9479 0.850
White’s test 𝐻0: 𝐻𝑜𝑚𝑜𝑠𝑘𝑒𝑑𝑎𝑠𝑡𝑖𝑐𝑖𝑡𝑦
𝐻1: 𝑈𝑛𝑟𝑒𝑠𝑡𝑟𝑖𝑐𝑡𝑒𝑑 ℎ𝑒𝑡𝑒𝑟𝑜𝑠𝑘𝑒𝑑𝑎𝑠𝑡𝑖𝑐𝑖𝑡𝑦
Chi2 Prob > Chi2
