¨
The impact of quantitative easing on firm financing
A Swedish studyAuthor: Alexander Fredriksson (931024)
Spring 2019
Master thesis in Finance, Advanced level, 30 credits Master’s in finance
Örebro University School of Business Supervisor: Pär Österholm
Abstract
Slow growth and low inflation has prompted central banks in the western economies to use unconventional forms of monetary policy. This has primarily consisted of what is referred to as Quantitative Easing, where the central bank carries out large scale purchases of financial assets. In order to improve the understanding of these policies economic influence, this thesis sets out to study its effect on Swedish firm’s financial decisions regarding leverage, capital expenditure and shareholder yield. A unique dataset of quarterly observations from 2006 to 2019 of 316 publicly traded firms is compiled and tested with panel regressions. The findings are overall inconclusive compared with previous studies. There are no indications of increased leverage or investments, but there are some hints at eventual income distributive effects through increased shareholder distribution.
Contents
1. Introduction ... 1 2. Theory ... 3 2.1 Transmission Channels ... 3 2.2 Capital Structure ... 5 2.3 Capital Expenditure ... 52.4 Dividends and Buybacks ... 7
3. Literature Review ... 9 3.1 Capital Allocation ... 9 3.2 Quantitative Easing ... 11 4. Data... 12 5. Methodology ... 15 5.1 Leverage ... 16 5.2 Capital Expenditure ... 17 5.3 Shareholder Yield ... 18 6. Results ... 18 6.1 Leverage ... 19 6.2 Capital Expenditure ... 21 6.3 Shareholder Yield ... 22 6.4 Sector Comparison ... 23 7. Sensitivity Analysis ... 24 7.1 Outliers ... 24 7.2 Observation Frequency ... 25 8. Discussion ... 26 8.1 Interpretation ... 26
8.2 Limitations and Validity ... 28
9. Conclusion ... 29
10. References ... 30
1
1. Introduction
Following the financial crisis in 2008 the Riksbank has utilized Quantitative Easing-programs (QE) to stimulate the Swedish economy and improve financial stability. The Swedish economy has, similarly to most developed countries, had a drawn-out problem with a low level of inflation combined with slow growth. Central banks in these countries have pushed the interest rate very low, with the Riksbank even having a negative interest rate from February 2015 to January 2020. With this measure not having enough of a desired effect QE was also introduced in 2015 as an attempt at further economic stimulation.
QE is a form of expansionary monetary policy where the central bank carries out large scale purchases of government bonds and other financial assets. These operations stimulate the economy through several transmission channels including injecting liquidity into the economy, sending positive signals to investors about the future commitments of the central bank, shifting the portfolios of investors into riskier assets, lowering the real interest rate and putting downward pressure on the exchange rate. This in turn leads to increased economic activity by increasing lending and investments. The Swedish development can be observed in figure 1 below. Beginning in 2015 there is a steady rise in the Riksbank’s asset levels combined with a reduction in the repo rate to historically low levels, and their divergence shows an increase in expansive monetary policy used to stimulate the Swedish economy.
Figure 1. Riksbank’s total assets and the repo rate, 2006-2019.
Note: Figure showing the Riksbank’s total assets in million SEK (left axis) and the repo rate (right axis). Quarterly data, 2006-2019. -1,00 0,00 1,00 2,00 3,00 4,00 5,00 0 200000 400000 600000 800000 1000000
2
As a result of these programs the balance sheet of the Swedish central bank has increased fourfold during the period following the financial crisis, now equaling about 20% of Swedish GDP at 900 billion SEK (Riksbanken, 2020). As QE has historically not been a standard tool of western central banks, much of its effects are still being studied. While some find evidence of a positive impact from these operations (Darracq and De Santis, 2015; Cohen, Gómez-Puig &, Sosvilla-Rivero, 2019) some dispute its effectiveness altogether (Thornton 2012). Yet others have raised points about its potential influence on income inequality (Coibion, Gorodnichenko, Kueng, & Silva, 2017) As such, the purpose of this essay is to analyze to what extent the central banks QE-program has influenced Swedish firm’s leverage and capital allocation decisions. More specifically the aim is to examine if the actions of the Riksbank has encouraged firms to raise their debt level, expand their investments and increase their shareholder distribution. To accomplish this, the three variables leverage, capital expenditure and shareholder yield are used as measurements of capital allocation. Panel regressions are used to test if changes in the Riksbank’s assets have a statistically significant influence on the three variables. The results are overall inconclusive, with some findings hinting at eventual income distributive effects.
This thesis has the following structure: Section 2 contains an overview of related theory. Section 3 contains a literature review. Section 4 summarizes the data used. Section 5 covers the methodology used for the study. Section 6 has the results of the panel data regression. Section 7 contains a sensitivity analysis of the results. Section 8 covers the discussion of the results and section 9 concludes.
3
2. Theory
This section contains an overview of the theory that relates to this thesis. The first part consists of an overview of the transmission channels of QE. This is followed by a review of capital allocation theory divided into subsections for capital structure, capital expenditure and shareholder distribution respectively.
2.1 Transmission channels
Interest rate channel
The traditional interest rate channel stems from the classic Keynesian IS-LM model for monetary transmission which states that expansionary monetary policy causes real interest rates to fall. This then lowers the cost of capital which in turn increases investment spending and aggregate demand. Keynes identified this channel as primarily operating through businesses investment decisions, but later research has also included consumers housing and consumer durable expenditure investment decisions (Mishkin, 1996). When the central bank alters the term nominal interest rate, the effect of sticky prices results in a reduction of the short-term real interest rate as well. According to the expectations hypothesis, long-short-term interest rates are an average of expected future short-term interest rates provided that bonds with differing maturities can be perfect substitutes to each other. As such, the lower real short-term rates in turn affect the real long-term interest rates. These then drive up business and consumer investments.
Signaling channel
One of the main factors that impact long-term interest rates is the expectations about future short-term interest rates. QE can be utilized to influence the market participants expectations. When the central bank announces its intention to keep short-term interest rates low QE helps increase the credibility of the central bank’s commitment. Taking an early exit from the strategy would result in losses, which indicates that the policy will be in place for a longer period of time (Gern et al., 2015).
Portfolio rebalancing channel
When long-term and short-term bonds are not perfect substitutes their relative supply impacts the yield curve. If the central bank moves long-term government bonds from the market to its balance sheet, the term premium of these bonds is reduced. Investors are then forced to
4
rebalance their portfolios, partly by moving into riskier assets with a higher yield (Gern et al., 2015). Central banks can also buy private sector assets such as corporate bonds and mortgage-backed securities. As these assets are riskier investments than government bonds, they act to reduce the market risk premiums.
In a perfect market with rational investors the supply of a security should not be able to affect its price, based on the CAPM and arbitrage-free model (Williams, 2011). To account for the possibility of a portfolio rebalancing channel, models that account for heterogeneous investor preferences, such as the preferred habitat theory of Modigliani and Sutch (1966), are needed. It is hard to disentangle the effect from a potential signaling channel and portfolio rebalancing channel, but the signaling channel would cause a strong co-movement for all long-term yield asset classes. The portfolio channel indicates a relatively muted response for all asset classes that are not close substitutes to government securities. There are some empirical results that indicate the possibility of an existing portfolio rebalance channel (Williams, 2011).
Credit channel
If the central bank purchases government securities from the banking sector they expand their balance sheet while in turn increasing the liquidity in the banking system. This in turn increases the cash reserves of the banks. When the reserves reach a high enough level, this effect will eventually continue into the broader economy through bank lending operations. This drives up asset prices and counteracts deflation (Joyce et al., 2012).
Exchange rate channel
The main way that QE affects the exchange rate is through the reduction in interest rate that follows from expansionary monetary policy. As the interest rate falls, its value relative to other currencies is diminished. Findings by Glick and Leduc (2013) show that surprise QE announcements are followed by a strong decline in the value of the dollar compared to several currencies. Changes in the relative interest rate does not appear to fully explain the exchange rate effect however. Haldane et al. (2016) find that only about half of the currency depreciation could be explained by the interest rate. QE might also have an effect on long-term exchange rate expectations. Announcements about monetary expansion could indicate higher uncertainty and increase the foreign exchange risk premium.
5
2.2 Capital Structure
The theoretical framework for capital structure theory have its roots in Modigliani and Miller
(1958) who presented the theory that in a full-information, tax-free world capital structure would have no effect on a firms value. If these market imperfections don’t exist a firm would solely choose its leverage ratio based on cost of capital and available financing. Factors such as tax shields and principal-agent relations end up affecting firm value and in turn influence firm’s capital decisions.
Literature that followed tried to create a theory for how firms carry out their capital allocation which ended with two main theories emerging. The first one was proposed by Kraus and Litzenberger (1973) who presented the trade-off theory of an optimal capital structure for each individual firm. It is dependent on the cost and benefit of altering the debt equity ratio, where higher debt increases the tax shield but raises the financial distress cost. As such firms depend on not just their own situation but are dependent on external factors such as interest rates in finding its target leverage ratio.
The second major theory to emerge was the pecking order theory by Myers (1984) which states that firms will choose their capital structure on the basis of preferring internal funds before external financing, and debt before equity. This is based on the idea that investors with incomplete information will take firm capital structure choice as a signal of the management’s belief in the firm’s performance. Taking on debt signals that management believes the firm can take on more risk and relying on equity is regarded as a last resort measure when all other options are exhausted.
2.3 Capital Expenditure
Tobin (1969) concludes that the principal way that financial policies can impact aggregate demand is by changing the value of physical assets relative to their replacement cost. Changes in monetary policy can also impact the portfolio preferences by shifting the asset demand function. This reasoning is the foundation for the portfolio rebalancing theory.
Fama & Gibbons (1982) find empirical evidence for the hypothesis that the expected real return of interest rate has a negative correlation with expected inflation. In contrast to previous studies they concluded that this was an outcome primarily from the capital expenditure process. Their findings show that expected real returns vary with capital expenditure to induce equilibrium allocation between consumption and investment. In conjunction with this they find a negative relationship between expected inflation and real activity which they conclude is an effect from
6
the monetary sector. This results in a negative relationship correlation between expected inflation and real returns. The process is explained as expected inflation raising the opportunity cost of money which causes investors to shift to interest-bearing assets, this in turn depresses real returns on these assets. Lower expected real returns then reduce the cost-of-capital for capital expenditure which results in increased investments.
In their study McConnell and Muscarella (1985) study public capital expenditure announcements of 658 firms between 1975 and 1981 to test the market value maximization hypothesis. This states that managers will invest to the point where the marginal rate of return on invested funds equals the market required rate of return, which means that unexpected increases in capital expenditure should cause the market valuation to increase and unexpected decreases should cause it to decrease. It is contrasted to the size maximization hypothesis, where managers are believed to want to maximize the firm size which results in capital overinvestment. If this is the prevalent behaviour the relationship between capital expenditure and market valuation should be reversed, where unexpected increases cause the valuation to decrease and vice versa. With their results they conclude that it supports the market value maximization hypothesis, which means that managers make their capital expenditure decisions consistent with attempting to maximize firm valuation over size.
In their study Blundell-Wignall and Roulet (2013) present empirical evidence of the cost of equity drifting upwards in the US since 2000 from around 8% to 10%. This development is combined with the cost of corporate debt trending strongly downwards during the same time period. In this period the high-yield bond in the US has fallen below the earnings yield on equities in the S&P 500. The authors present evidence for the risk premium rising substantially due to the increased difference in the cost of debt and cost of equity. Due to this development a very strong incentive has arisen to issue debt and retire equity. The authors present the equity risk premium as:
𝜎 = (𝑑 + 𝑔) ∗ (1 − 𝑐) − 𝑖 ∗ (1 − 𝑡) 1
Where 𝜎 is the equity risk premium. 𝑑 is the company dividend yield, 𝑔 the ex-dividend nominal return for a dollar invest, 𝑐 is the capital gains tax, 𝑖 is the interest rate on corporate debt and 𝑡 is the personal tax rate. They argue that with the current economic climate of such a high-risk premium there are very low incentives to carry out investment. The very low interest rate instead incentivizes an increase in borrowing with the intention of carrying out buybacks.
7
2.4 Dividends and Buybacks
Black & Scholes (1974) pioneered research on the impact of dividend policy on share prices. They built upon the idea by Modigliani and Miller (1961) who proposed that over time any firm would attract to itself a clientele that preferred its particular dividend policy. If investors had differing preferences about dividend policy firms could then differentiate their dividend policy to meet their demand. At a certain point an equilibrium would be reached however, where the firms will have supplied the investors with all potential dividend policies that they seek. Changing the dividend policy of an individual firm would then just shift the clientele between firms, but not alter the valuation of any specific firm. Only a short-term uncertainty effect would be prevalent, where increasing or lowering dividends could signal positive or negative changes in expectation, but the valuation would return to normality over time as these expectations are not fulfilled. Black and Scholes (1974) provide empirical results that indicate this hypothesis hold true, and there are no effects of dividend policy on stock returns.
Fairchild (2006) attempts to explain why firms carry out share buybacks. He begins by rejecting the notion that a repurchase can increase the share price of a firm in a perfect market. To illustrate this a firm with a market value of $5 000 000 and 100 000 shares outstanding is used as an example. The firm has made a profit of $1 000 000 this year, which is similar to last year, and decide to utilize it for repurchases instead of dividends or investments. This gives a value of equity as:
𝑉 = 1 000 000 +1 000 000
ρ = 5 000 000 2
It gives the implied cost of equity ρ of 25% which is the investors required return for investing in the company. If the firm has no positive net present value (NPV) projects to invest in it has three different courses of action to choose from. It can choose to invest in a zero NPV project, give the profits as a cash dividend or use it for share repurchase. If the firm chooses to reinvest the profits in a zero NPV project the impact of the firm value is:
𝑉 = 1 000 000 +1 000 000
0.25 = 5 000 000 3
This means that the value per share remains at $50 per share. Should the firm instead choose to give the profits out as dividends the impact on firm value is:
𝑉 =1 000 000
8
Which means that the value per share is now $40 per share. This does not mean that the wealth for the investor is different from the first example, however. The reduction in share price has been exactly offset by the cash dividend of $10 per share, so the investors are exactly as well off as in the first example with a shareholder wealth of $5 000 000. Finally, the firm can choose to use the profits for a share buyback. If the firm’s initial value is:
𝑉 = 1 000 000 +1 000 000
0.25 = 5 000 000 5
Using the $1 000 000 for a cash buyback results in 20 000 shares being purchased, with 80 000 remaining outstanding. The new firm value is:
𝑉 =1 000 000
0.25 = 4 000 000 6
With the new value of $4 000 000 divided by the lower number of shares of 80 000 the price per share still equals $50. The $1 000 000 that were used for repurchasing were given to shareholders, so the total shareholder wealth still remains at $5 000 000. With this Fairchild (2006) proves that with no market imperfections and no positive NPV projects available the firm’s decision has no impact on its share price. So why do firms utilize share buybacks? To answer this question Fairchild (2006) raises three potential factors that can explain why share repurchasing can impact firm value. These are referred to as the signalling motive, the timing motive and the capital structure motive. The signalling motive is that repurchasing can be utilized as a tool to signal that the shares are undervalued which results in a price increase. The timing motive is based on the idea that if the firm is undervalued management can use share repurchases to create shareholder wealth, as they would be buying the shares under market value. Neither of these can arise if there are no market imperfections, such as information inefficiency or lack of investor rationality. The capital structure motive has its impact through managerial motivations. Outside capital dilutes the managers own equity stake in the company so shareholder buybacks can serve as a way to reduce outside capital. Yet another factor is the combination of increased leveraging with buybacks, which raises the firm’s value through increasing the tax shield.
Cohen et al. (2019) show that if a firm has the budget constraint of:
9
where 𝐼 is resources devoted to investment and 𝑉 is paying shareholders via dividends or buybacks, a solution can be derived for the firm’s maximization problem for capital allocation such as:
π´(I)
(1 + r)= 𝜌 8
Here π´(I) is the first order condition for investment, r is the company’s cost of debt and 𝜌 is the company’s return to shareholders. Given this solution a firm will split its resources into investment and shareholder distribution until the point where the discounted marginal return on a project equals the added value a dividend or buyback has on the firm’s stock price.
3.
Literature Review
This section contains an overview of previous studies that relate to the thesis. The first part covers empirical studies on capital allocation behavior and the second covers empirical findings on the economic effects of QE.
3.1 Capital Allocation
Titman and Wessels (1988) perform an empirical study on capital structure. They use a factor-analysis technique to estimate unobservable attributes effect on choice of corporate debt ratio. A major implication of their findings is that the composition of debt instruments that make out the firm’s debt structure plays a significant role in their behavior. Long-term, short-term and convertible debt expose the firms to different risks. In addition, they find a correlation between debt composition and firm characteristics, such as size and sector.
To expand on the understanding of macroeconomic factors on firm behavior, Halling, Yu and Zechner (2016) study the impact of the business cycle on firm leverage. Their main finding is that firm’s target and realized leverage evolves counter-cyclically to the business cycle. That is, during a recession firm leverage increases and during a boom it decreases, despite the cost of borrowing developing pro-cyclically. They explain this as the present value of future earnings increasing during a recession which offsets the increased cost of borrowing. Another important finding is the importance of financial constraints on firm leverage behavior. Financially constrained firms are more hampered in their ability to adjust their leverage, which becomes important to account for in empirical studies of leverage.
10
Kühnhausen & Stieber (2014) use a panel data set of a large number of firms to compare differences in leverage between non-financial firms. They study the specific firm characteristics in determining leverage, and compare them between countries, industries and size. Their main finding is that firm size, industry leverage, industry growth and tax shield all have a positive impact on firm leverage while profitability and liquidity have a negative impact. As previous literature has found a significant difference in how these characteristics can vary between countries, they test for and find the highest average leverage ratios in Germany, Italy, Monaco and Norway, while the countries with the lowest leverage ratios are Turkey, Ukraine and the US. There are also variations in how the firm specific characteristics impact leverage ratios in the different countries. When testing for firm size larger firms have a higher level of leverage ratios compared to small firms and the results indicate that large firms are more impacted by macroeconomic factors than small firms. They also test for several macroeconomic factors such as inflation, GDP growth, capital flows, unemployment and stock prices, but the findings are more inconclusive and vary between countries.
Denis and Osobov (2008) study changes in firm dividend policy. They build upon the findings of Fama and French (2001) who find that dividends in the US have been concentrated to a smaller number of high paying firms. Denis and Osobor (2008) study a number of developed nations to compare the development of dividend policy to that in the US. Their findings do not produce enough empirical evidence to conclude that there are larger changes in dividend policy for the period 1989-2002, and the changes that they do observe are due to smaller firms who do not pay dividends entering the market, instead of larger firms altering their dividend policy. Their findings show that the propensity to pay dividends depend primarily on firm size, growth opportunities and profitability. More importantly they find that it depends heavily on ratio of retained earnings to total equity. They show a positive correlation between a high level of retained earnings and dividend pay-outs. These findings go against the theory of signalling and clientele as the primary determinants of dividend policies. Instead their findings produce support for the life-cycle theory. This states that firms will alter their dividend pay-outs in development with their investment opportunity set. Its prediction is that firms will have low dividends in their infancy as their investment opportunities exceed their internally generaed capital. As the firm matures its internally generated capital will surpass its investment opportunities, resulting in the firm contributing more of its funds to shareholder distribution as not to waste the free cash flow.
11
3.2 Quantitative Easing
Darracq-Paries and De Santis (2015) study the impact of the ECBs Long-Term Refinancing Operations (LTRO) which were introduced in 2011 and 2012. These were 3-year loans used as liquidity provisions and amounted to 489 billion euro in 2011 and 530 billion euro in 2012 with an interest rate at 1% and equalled 10.8% of the euro area nominal GDP. Isolating the credit supply shock invoked by this measure the findings of the authors indicate that it had a positive impact on GDP and loan provisions to non-financial corporations. Results indicate that the effect of loans was the largest of the two at a 2% increase, compared to GDP which had a 0.7% increase. The peak of the credit supply transmission effect was achieved after 5 quarters and the loan impact was at its highest point 10 quarters after initiation. In their study the authors find indications that the programs had a lagging effect on the business cycle which they speculate could be due to debt being rolled over.
Some skepticism exists in regards to the effectiveness of QE as a stimulus tool. Thornton (2012) argues that the portfolio balance channel, considered to be one of the main drivers for economic stimulus, has weak evidence behind it. He considers the habitat model (Modgliani & Sutch ,1966) to be underutilized in analyzing investor behavior connected to the channel. If long-term secure bonds are removed from the market, risk averse investors will not shift into riskier assets but instead attempt to find other low risk assets. An example used is the shift of pension funds in investing into more real estate following a large drop in the long-term yields in the UK in 2004. An empirical study is done on interest rates and public debt supply, but the author finds no evidence that declines in long-term rates are due to the portfolio rebalance channel.
To bridge the gap between studies on capital allocation and economic effects of QE, Cohen et al. (2019) use panel data regression on firms active in the Economic and Monetary Union (EMU) to study the effects of QE on firm’s capital allocation decision. They rely on changes in the ECBs assets as a proxy for QE and the Euribor rate as a proxy for the central banks impact on the interest rates. Their findings show that the expansive monetary policy of the ECB leads firms to increase their leverage, capital expenditure and shareholder distribution. A comparison of countries and sectors show that the effect of the response is homogenous across them, but its magnitude vary between the samples. Their conclusion is that the expansive monetary policy of the ECB not only leads to increased economic activity but has certain distributional effects as well.
12
4. Data
This section presents the dataset for the study. In addition, an overview of the presence of missing observations amongst the independent variables and the procedure for handling extreme outliers in the data are included.
Like Cohen et al (2019) financial data has been gathered from Bloomberg. The chosen firms are publicly traded firms on the Stockholm exchange (OMX). In total, data is gathered for 316 firms excluding 15 additional firms due to data unavailability. The timespan is 56 quarters, from the first quarter of 2006 to the fourth quarter of 2019. The three dependent variables that are studied are Equity, Capital Expenditure-to-Sales and Shareholder Yield. Debt-to-Equity is the summation of short-term debt, long term debt and other fixed payments, all divided by shareholder equity. Capital expenditure-to-sales is all capital investments divided by total sales. Shareholder yield is the summation of dividends, net share repurchases and net debt repayments divided by market capitalization.
A firm has three main choices in regards to its yearly profits. It can distribute them directly to the shareholders as dividends, utilize them for share repurchases and debt repayments that in turn benefit shareholders as capital gains or use them for reinvestments to expand the firm. Shareholder yield is used to capture all forms of shareholder distribution, which means all profits that are not reinvested back into the firm. Since most Swedish firms only carry out yearly dividends, but market capitalization changes throughout the year, the effect on shareholder yield can potentially become obfuscated by changes in market capitalization instead of returns to shareholders, this is covered further in the robustness analysis and discussion. For firm specific independent variables EBITDA (Earnings before interest, taxes, depreciation and amortization) to sales, growth in EPS (Earnings per Share) and WACC (Weighted Average Cost of Capital) are used. The descriptive statistics for the gathered data can be viewed in table 1 below.
13
Table 1. Observations, mean, standard deviation and min/max values for gathered data
Variable Observations Mean Standard Dev. Min Max
Leverage 11,897 73.53 117.33 0 1049 Capital Exp 7,434 20.23 58.31 0.014 587.15 Shareholder Yield 5,930 -2.03 14.10 -68.73 48.61 Riksbank Assets 56 546963 259069 189459 951360 Stibor 3M 56 1.268 1.643 -0.568 5.179 Inflation 56 1.5 0.695 0 3.5 High-Low Spread 56 0.774 0.914 -1.046 2.793 EBITDA-to-Sales 10,706 -9.84 171.72 -2399 116.55 EPS Growth 11.145 36.05 160.52 -605.62 1566.67 WACC 12.982 7.11 2.48 1.22 14.85
Note: Table of descriptive statistics for the gathered data after outliers have been removed. Includes observations, mean, standard deviation and min/max values.
The variables that are used to control for monetary policy are changes in the assets of the Riksbank and the weighted 3-month Stibor rate (Stockholm Interbank Offered Rate), which are both gathered from the Riksbank. Changes in the Riksbank’s assets are used as an overall proxy for QE. The Stibor rate is used as a proxy for the central banks effect on interest rates. It is utilized instead of the repo rate as it is believed to be more representative of the interest rate that firms actually face. The spread between the 10-year sovereign bond and 3-month Stibor rate is used as a proxy for economic conditions, while the inflation rate is used as a proxy for economic activity. Inflation rate is measured as the CPIF (consumer price index with a fixed interest rate). Data for the high-low spread is gathered from the Riksbank while data for inflation is taken from SCB. Graphs of the data for all the macroeconomic variables can be viewed in figure 2 below.
14
Figure 2. Graphs of the Riksbank’s asset levels, 3-month Stibor rate, inflation rate and the 10 year bond/3-month Stibor spread, 2006-2019.
Note: Graphs showing the macroeconomic variables Riksbank’s asset levels (top left), 3-month Stibor rate (top right), inflation rate (bottom left) and the 10 year bond/3-month Stibor spread (bottom right). Quarterly observations, 2006-2019.
A number of extreme outliers can be observed in the original dataset. As the data from Bloomberg is self-reported, several of these are can be due to data errors and their magnitude risk skewing the results of the regression. To handle this issue the top and bottom percentile of observations for the variables leverage, capital expenditure, shareholder yield, EBITDA, EPS and WACC have been removed. The data in table 1 is after the removal of outliers. Table 2. Percentage of missing values for variables over different time periods.
2006Q1-2019Q4 2009Q1-2019Q4 2012Q1-2019Q4 2015Q1- 2019Q4 Leverage 32.56% 24.86% 17.97% 9.87% Capital Exp 57.86% 51.70% 44.39% 34.41% Shareholder Yield 66.38% 58.15% 49.35% 36.51% EBITDA-to-Sales 39.31% 34.09% 28.39% 21.35% EPS Growth 36.82% 30.24% 24.22% 16.86% WACC 26.41% 21.88% 16.73% 10.21%
Note: Table of percentage of missing values for the entire sample (2006Q1-2019Q4) and three subsamples (2009Q1-20019Q4, 2012Q1-2019Q4, 2015Q1-2019Q4) after the removal of outliers.
15
Due to the nature of self-reported data and some firms getting listed during the studied period there are missing values that vary for both the dependent and independent variables. As can be observed in table 2 above, leverage has the least amount of missing values of the dependent variables with 32.56%, compared to capital expenditure at 57.86% and shareholder yield at 66.38%. The prevalence of missing values is reduced over time, so the amount of missing variables falls to 9.87% for leverage, 34.41% for capital expenditure and 36.51% for shareholder yield. There are no breaks in the data for the panels. This means that all firms that have reported shareholder yield for Q1 2006 continue to report over the whole period until Q4 2019. Those that begin to report it for Q2 2006 continue for the whole period, and so on. The panel unspecific data, which is the Riksbank's assets, the 3-month Stibor rate, inflation and the 10-year/3-month bond spread, are excluded from the analysis as they have no missing values.
5. Methodology
The econometric model used is this thesis is the same as the one used by Cohen et al. (2019) to test the effect of expansive monetary policy on firms capital allocation. In their study they found empirical evidence for an increase in the ECBs assets and a reduction in interest rates having a significantly positive impact on firm’s capital expenditure, leverage and shareholder yield. To achieve this, they relied on individual panel regressions for all three variables. Like in their process, a Hausman test will be used to find out if a Fixed-effects (FE) model or Random Effects (RE) model is the better fit. To be able to compare if the effect could be different after the Riksbank began to expand its balance sheet in 2015 the regressions are also carried out in two subsamples based on time periods. This is intended to provide separate coefficients for comparison. The first period is from quarter 1 2006 to quarter 4 2014 and the second is from quarter 1 2015 to quarter 4 2019, which means the second period encapsulates the period with QE.
To study the effect of monetary policy in different sectors the regressions are also done on subsample. These sectors are classified according to the one used by OMX Stockholm and contains Industrials, Consumer Goods, Consumer Services, Basic Materials, Health Care, Telecom, Financials, Technology and Oil & Gas. The Utilities sector is disregarded as there is only a single firm from the sector present in the dataset. A limitation of this comparison is that applying the same model with the same explanatory variables across different subsamples can result in control variables that are only relevant for the whole sample being included for the
16
subsample. This can lead to higher variance and an erroneous rejection of statistical significance (Cohen, 1983). Excluding the nonsignificant variables from the subsample would instead risk removing important variables and creating biased estimates. Therefore, the same model is applied to all subsamples but this limitation is important to keep in mind when interpreting the results.
When doing panel regressions, it is important to account for unit roots in the data. A unit root in a time series is a stochastic non-stationary process that follows an unpredictable pattern. If there are unit roots present in the data it can give rise to so called spurious regression. These will result in variables being correlated as they both vary dependent on time and can result in statistical significance being shown in a regression when there is no causal relationship (Granger and Newbold, 1974). To test if unit roots are present in the data, the Fisher-type augmented Dickey-Fuller and Philips-Perron tests are applied to the firm specific variables and the standard augmented Dickey-Fuller and Philips-Perron tests are used on the univariate variables. The results, which can be observed in table 6 in the appendix, show the presence of unit roots for the Riksbank’s assets. In order to handle the presence of these unit roots the first difference of the variable is used, and following tests indicate first-difference stationarity. Both Stibor and the high-low spread have different results dependent on test. As stationarity can be rejected in the augmented Dickey-Fuller test and their level effect in itself is believed to be relevant in the model, they are kept in their level form.
Like Cohen et al. (2019) three equations have been estimated. One for leverage, one for capital expenditure and one for shareholder yield.
5.1 Leverage
𝐿𝑖,𝑡 = 𝛼𝑖,𝑡+ 𝛽1∗ 𝑋𝑖,𝑡−1+ 𝛽2 ∗ 𝑌𝑡−1+ 𝛽3∗ 𝑍𝑡−1+ 𝜀𝑖,𝑡 9 In equation 9 above, the model has leverage (𝐿𝑖,𝑡) as the independent variable which is measured by the firm’s debt to equity. The model has three independent vectors which are the X-vector, Y-vector and Z-vector. In the X-vector the three firm specific microeconomic variables EBITDA, growth in EPS and WACC are included. EBITDA and EPS are included since they are believed to play a big role in a company’s leverage ratio. WACC is included as it has a major impact on a firm's leverage and a negative relationship is expected.
17
The Y-vector contains inflation rate as a separate macroeconomic factor used to measure economic activity. Its expected relationship with leverage is not entirely evident. Increased inflation can result in higher leverage due to its depreciative effect on debt levels measured in real terms. Simultaneously, higher inflation leads to increases in the real interest rate which in turn has a negative effect on leverage ratios. As such, the expected relationship is considered ambiguous.
In the Z-vector the monetary policy variables are included. These are the change in the Riksbank’s assets, serving as a proxy variable for the QE-programs, and the level of the 3-month Stibor rate. A lower interest rate should reduce the cost of borrowing resulting in an increased level of debt, so a negative effect is expected from the interest rate. Usage of the QE-program should have the same effect, which means a positive effect is expected between the Riksbank’s asset level and leverage.
5.2 Capital Expenditure
𝐶𝑖,𝑡 = 𝛼𝑖,𝑡+ 𝛽1∗ 𝑖𝑡−1+ 𝛽2∗ 𝑅𝐾𝐵𝑡−1+ 𝛽3 ∗ 𝑆𝑡−1+ 𝛽4∗ 𝑃𝑡−1 10 +𝛽5∗ 𝐸𝑖,𝑡−1+ 𝛽6∗ 𝑘𝑖,𝑡−1+ 𝜀𝑖,𝑡
In the above equation the dependent variable is Capital expenditure divided by sales (𝐶𝑖,𝑡). The model also includes the same two monetary policy variables, the 3-month Stibor rate (𝑖𝑡−1) and changes in the Riksbank’s assets (𝑅𝐾𝐵𝑡−1). The remaining four independent variables are the
inflation rate (𝑃𝑡−1), the cost of capital in the form of WACC (𝑘𝑖,𝑡−1), EBITDA (𝐸𝑖,𝑡−1) and the long-term and short-term yield spread (𝑆𝑡−1) measured as the difference between a 10-year sovereign bond and the 3-month Stibor rate.
As in equation 9 changes in the Riksbank’s assets and the cost of debt, measured as the 3-month Stibor rate, are included as the main central bank policy variables. The WACC is included to test for the cost of capital and is expected to have a negative relationship as an increase in cost of capital reduces returns on investments. EBITDA is included as a measure of changes in profitability and is expected to have a positive effect on capital expenditure. Inflation rate and the long-term/short-term yield spread are included to measure changes in the economy and market expectations. The expected relationship between capital expenditure and inflation rate is not entirely clear. It can have both a negative effect by lowering the real returns
18
and thus reducing profitability, and a positive effect as a rise in the inflation rate indicates higher economic activity. The spread of long-term/short-term rates is expected to be positive, in line with the reasoning of Baumeister and Benati (2013) who argue that a compression of the spread may impact both GDP and inflation while indicating a fall in the term premium. Contraction of the spread thus indicates that firms might be reducing their capital spending due to negative developments in economic activity.
5.3 Shareholder Yield
𝑦𝑖,𝑡 = 𝛾𝑖,𝑡 + 𝛽1∗ 𝑖𝑡−1 + 𝛽2∗ 𝑅𝐾𝐵𝑡−1+ 𝛽3∗ 𝐸𝑖,𝑡−1+ 𝛽4∗ 𝑘𝑖,𝑡−1+ 𝜗𝑖,𝑡 11
In the last equation Shareholder Yield is the independent variable (𝑦𝑖,𝑡). Similar to equation 9 and 10 it contains the same independent variables for monetary policy, changes in the Riksbank’s assets (𝑅𝐾𝐵𝑡−1) and the 3-month Stibor rate (𝑖𝑡−1). This model only includes two other independent variables, the cost of capital in the form of WACC (𝑘𝑖,𝑡−1) and change in profits measured by growth in EPS (𝐸𝑖,𝑡−1).
The WACC is expected to have a positive relationship with shareholder yield, as if the cost of retaining earnings to invest relative to the cost of bonds increase the firm has an increased incentive to repurchase shares. For changes in profits, the relationship is expected to be positive as an increase in earnings should result in increased returns to investors.
6. Results
To find if a fixed-effects or random effects method is the best specification for the model the Hausman test is used on the regressions for all samples. The test is used to detect if there are endogenous variables, variables that are determined by other variables in the regression, and if so the assumptions of ordinary least square as the optimal predictor is no longer true (Hausman, 1978). The null hypothesis is that endogenous variables are absent and the random effects model is the best fit. If it is rejected the fixed-effects method is used to counteract the issue. The results are presented in the results tables of the regressions. For leverage and shareholder yield the findings are that the fixed-effects method is the best fit, regardless of subsamples. For capital expenditure the fixed-effects method is the best fit for the sample of the entire time period and the later period, but the random effects method is better for the early time period.
19
For comparison between subsamples, the fixed-effects model is applied to all samples since it was the best fit for 2 out of 3.
6.1 Leverage
Table 3. Results of the fixed-effect panel regressions for leverage.
Variable 2006Q1-2019Q4 2006Q1-2014Q4 2015Q1-2019Q4 D(Assets(t-1)) -0.027 -0.022 0.033 0.0107** 0.0098* 0.0452 3M Yld(t-1) 1.87 2.79 56.39 0.471** 0.67** 5.28** EPS(t-1) -0.0001 -0.0020 0.0041 0.0037 0.0043 0.0053 WACC -4.459 -3.925 -4.604 0.338** 0.442** 0.492** EBITDA(t-1) 0.00006 -0.0134 -0.0016 0.0053 0.0074 0.0067 Inflation(t-1) 3.479 1.206 1.734 0.877** 1.078 1.734** Constant 87.45 81.75 83.99 2.657** 3.567** 4.208** R-squared 25.3% 9.71% 4.35% F-statistic 31.11** 16.70** 27.40** Total Obs. 8232 3991 3895 Hausman 60.70** 15.56** 160.42** RE/FE FE FE FE
Note: Table of results for estimation of the leverage regression.* and ** indicate statistical significance at the 5% and 1% level respectively. Statistical significance for Hausman test indicates fixed-effect model is used.
The results of the panel regression for leverage is shown in table 3 above. According to the Hausman test the fixed-effect model was the best fit for all models. Mixed indications can be observed for the central bank variables dependent on time period. For the entire period the findings show that a statistically significant effect for both changes in the Riksbank’s assets and the level of the 3-month Stibor rate can be observed at the 1% significance level. An increase in the change of central bank assets of 1 billion SEK reduces the debt-to-equity ratio, all else equal, by an average of -0.027 percentage units. The Stibor rate shows a positive
20
relationship in the regression, with a one percentage increase of the rate giving rise to a 1.87 percentage point increase in the debt-to-equity ratio. Both these findings are out of line with the expected results. Comparing the coefficients of the first and second time periods, one can observe that assets have a negative statistically significant, effect of -0.022 percentage per billion SEK added to the Riksbank’s balance sheet before 2015Q1. The later period shows no statistical significance for the assets. The effect of the Stibor rate vary by several magnitudes between the time periods, where a one percentage increase in the rate gives a 2.79 percentage point increase in debt-to-equity before 2015Q1, compared with 56.39 percentage for the later time period which is an incredibly high and questionable estimate. Amongst the firm specific variables only the WACC shows a consistent statistically significant effect that is prevalent for all time periods, while EPS and EBITDA show a very high standard errors for all observed time samples. Inflation shows a positive impact on debt-to-equity but is only statistically significant in the period after 2015Q1. The R2 is the highest for the entire sample at 25.30%,
while the subsamples have notably smaller R2 with 9.71% for the early period and 4.35% for
21
6.2 Capital Expenditure
Table 4. Results of the fixed-effect panel regressions for capital expenditure-to-sales.
Variable 2006Q1-2019Q4 2006Q1-2014Q4 2015Q1-2019Q4 D(Assets(t-1)) -0.0169 -0.0211 0.0015 0.0079* 0.0092** 0.0316 3M Yld(t-1) 0.246 1.639 -5.748 0.406 0.874 5.125 Inflation(t-1) 0.075 0.053 -0.527 0.612 1.063 1.162 EBITDA(t-1) -0.0193 -0.0218 -0.0132 0.0034** 0.0058** 0.0045** High-Low(t-1) -0.679 0.156 -2.158 0.634 0.885 2.018 WACC(t-1) -0.222 -0.355 -0.017 0.241 0.433 0.341 Constant 15.92 13.58 16.59 2.02** 3.64** 3.43** R-squared 0.74% 1% 0.08% F-statistic 6.62** 3.92** 1.73 Total Obs. 6080 2561 3257 Hausman 22.80** 5.06 24.59** RE/FE FE FE FE
Note: Table of results for estimation of the capital expenditure regression. * and ** indicate statistical significance at the 5% and 1% level respectively. Statistical significance for Hausman test indicates fixed-effect model is used.
The results for the capital expenditure panel regression can be observed in table 4 above. The fixed-effect model was found to be the best fit for the overall and later sample, while the random effects model was a better fit for the early sample. To allow for comparison the fixed-effects model is used on all samples. An increase of 1 billion in the change of the Riksbank’s assets show a statistically significant negative effect on the capital expenditure-to-sales ratio of -0.0169 percentage points. Comparing the subsamples, the effect is larger before 2015Q1 at -0.0211 percentage, but statistically insignificant for the period after. The 3-month Stibor rate shows no statistically significant effect for any sample, which is unexpected. As with the results in the leverage equation, these findings are out of line with expectations.
22
For the firm specific variables, EBITDA-to-sales shows a statistically significant negative effect of -0.0193 percentage points. The effect is prevalent for both time periods, being slightly larger in the early period (-0.0218) and slightly smaller in the later period (-0.0132). WACC shows no statistically significant effect in any sample. Both the economic indicators inflation and high-low spread show a statistically insignificant effect both in the entire and the subsamples. The R2 is very low for all regressions, with 0.74% for the entire period, 8% for the first and 0.08% for the second period, which indicates the model is not a good fit to explain capital expenditure decisions.
6.3 Shareholder Yield
Table 5. Results of the fixed-effect panel regressions for shareholder yield.
Variable 2006Q1-2019Q4 2006Q1-2014Q4 2015Q1-2019Q4 D(Assets((t-1)) 0.0029 0.0012 0.0037 0.0038 0.004 0.0101 3M Yld(t-1) -0.485 -0.274 2.537 0.184** 0.315 1.290* EPS(t-1) -0.0007 -0.0017 -0.0008 0.0011 0.0017 0.0013 WACC(t-1) 0.428 0.488 0.127 0.094** 0.181** 0.317 Constant -4.516 -4.047 -2.991 0.658** 1.373** 0.909** R-squared 6.12% 13.14% 0.28% F-statistic 5.82** 2.18 2.02 Total Obs. 5361 1742 3382 Hausman 20.67** 14.88** 17.59** RE/FE FE FE FE
Note: Table of results for estimation of the shareholder yield regression.* and ** indicate statistical significance at the 5% and 1% level respectively. Statistical significance for Hausman test indicates fixed-effect model is used.
Results for the shareholder yield panel regression are shown in table 5. Similar to the leverage regression the fixed-effect model was found to be the best fit for all samples. In difference to the regressions for debt-to-equity and capital expenditure there is no statistically significant effect from the change in the Riksbank’s assets on shareholder yield for any sample. The 3-month Stibor rate shows a statistically significant negative effect for the overall sample,
23
however. An increase of 1 percentage point in the Stibor rate leads to a fall in the shareholder yield of -0.485 percentage. Observing the coefficients for the subsamples, the early period shows no statistically significant effect while the period after 2015Q1 has a statistically significant positive effect (2.537).
For the firm specific variables no statistically significant effect can be observed for the EPS in any sample. A positive and significant effect can be observed for the WACC at 0.428 percentage points for the entire sample and 0.488 percentage points for the early period. The R2 is once again very low at 6.12% for the entire sample, slightly higher at 13.14% for the early period and much smaller at 0.28% for the later period.
6.4 Sector Comparison
To observe if there are any significant differences in how monetary policy can affect different sectors, regressions are done for each sector individually which allows for unique coefficients to be estimated for each variable and sector. This method does not account for effects of interactions between the sectors. The output for these tests can be found in the appendix. The results for leverage vary over the sectors, with 4 out of 9 producing statistically significant coefficients from the effect of assets on leverage. The average effect varies between 0.036 percentage for telecom to 0.053 percentage for financial firms, but they all show a positive effect. Coefficients for Stibor vary as well, with Industrials showing the smallest effect at 3.615 and Consumer Goods having the largest effect at 11.84. A notable difference is the variation in R2 between the sectors which goes from 81% for Oil & Gas to only 1% for Consumer Goods while being on average around 25%.
The capital expenditure results show less statistically significant results within the sectors, where Industrials is the only one that shows an effect from assets. Industrials and Consumer Goods are also the only sectors who show a statistically significant effect for Stibor, but notably for Industrials the effect is positive (2.65) while for Consumer Good it is negative (-1.643). The R2 for these regressions are also markedly lower than for leverage, with only Telecom
(54%) and Technology (29%) above 10%.
Similarly to capital expenditure, the sector-individual regressions show very little statistical significance. Consumer services (0.021) is the only sector with a significant effect from changes in assets, which is notable however as the regressions for the entire sample showed no statistically significant results. Stibor only appears to have a significant effect on Consumer
24
Goods (-1.319) and Oil & Gas (-2.926). The R2 of these regressions are slightly higher and more varied than those for capital expenditure, with a variation from 1% for Health Care to 99% for Telecom but an average around 20% to 30%.
7. Sensitivity analysis
In this section additional tests are carried out to check the sensitivity of the reported results. The results of these tests can be found in the appendix. The first covers the handling of outliers in the data and the second covers the limitations of shareholder yield.
7.1 Outliers
To handle large outliers present in the data set the top and bottom percentile of the observations of firm specific variables were removed. In order to test if the method for handling outliers affects the results, the regressions are done once without removing any observations and once using the method of removing any observations that are 1.5 times the interquartile range above the 75th percentile and below the 25th percentile, referred to as the IQL-method. These can then be compared to the present findings to provide an overview of the sensitivity of the model to the handling of outliers.
Comparing the models for leverage, not removing any outliers from the dataset results in no significance for any variable except for inflation in the overall regression. There is also a large reduction in the R2, from 25.3% in the used model to 4.85% with all outliers included. Comparing this to the IQL method significance is attained for all variables except EPS. This is not unexpected, as removing large deviations will naturally reduce variance and standard errors. Additionally, the R2 is reduced to 18.61% despite more statistically significant variables.
This potentially hints at too many vital observations being removed with this method. The IQL-method results in only 6268 observations for the regression, while the used model has 8232 and the no outliers method results in 8810 observations. This is a total difference of 23.86% between the used model and the IQL-method, which gives an indication of a quite strong prevalence of deviating observations in the dataset.
For capital expenditure, the removal of no outliers results in significance only showing for EBITDA in the overall sample. The R2 of 6.51% is higher than the used samples minimal 0.74%. Applying the IQL-method on the sample instead only results in significance shown for the high-low spread variable and gives an R2 of 1.13%. Keeping all outliers gives a total of
25
6423 observations in the regression, compared with the 6080 in the used sample and 4846 with the IQL-method. Here the removal of outliers instead appears to primarily remove useful information, but none of the methods appear to explain why the explanatory ability of the model is low.
Keeping all outliers causes the results for shareholder yield to only show significance for EPS and gives a markedly larger coefficient estimate (-0.1088) compared to the used sample (-0.0007). It is possible that the presence of large outliers is distorting the other estimates in the regression. This hypothesis is supported by the large increase in R2 for the no outliers sample, 20.54% compared to 6.12% in the used sample, which appears to be driven primarily by the EPS. The IQL-method gives a sample with significance on the Stibor rate and WACC but has a smaller R2 at 1.16% and appears to once again remove too many vital observations.
Based on these results it appears that the outliers in the sample have a prominent effect and end up skewing the results of the regressions. Using the IQL-method instead however appears to remove outliers too aggressively. As such the used method of removing the top and bottom percentiles of observations might be the best trade-off between these two approaches, but the results are very sensitive to the choice between methods.
7.2
Observation Frequency
A limitation of this study is the dependent variable shareholder yield. The variable consists of the sum of dividends, deleveraging and share repurchases, divided by the market capitalization of the firm. As mentioned previously, most Swedish firms use primarily yearly dividends while the shareholder yield variable in this study is based on quarterly data. This can cause the denominator to change over the year while the numerator is primarily static, which can be perceived as changes to the shareholder yield that are not actually occurring. A possibility for dealing with this issue would be to run the regression on yearly observations instead, which reduces the total amount of time intervals from 56 to 14, and most critically, the time after the initiation of the Riksbank’s QE-programs in 2015 is reduced to 4 time intervals. To observe if there are marked effects from changes in the denominator, the regression is done on yearly data to compare to the results on quarterly data.
The yearly regression produces statistical significance for changes in assets, which was not present in the quarterly data. No other variable shows statistical significance, however. Most notably, the R2 is reduced from the already low 6.12% to 0.70% which indicates the model does
26
not explain more or less any variation in shareholder yield. These results indicate that while the results for the shareholder yield regression should be interpreted very carefully, running the regression on yearly observations does not appear to provide much information. Therefore, using the quarterly data, with its limitations, might be the most useful alternative.
8. Discussion
This section covers the discussion of the results. The first part discusses the interpretations of the results and how they relate to previous research. It is then followed by a discussion regarding the limitations and validity of the study.
8.1 Interpretation
The aim of this study is to test to what extent the Swedish central banks quantitative easing-program affects Swedish firm’s capital allocation decisions. The findings are mixed compared to previous studies on the subject. Changes in the assets of the Riksbank showed a negative effect on leverage which is different from the results that have been produced in previous studies, such as Cohen et al. (2019). One possible explanation might be the findings of Darracq-Paries and De Santis (2015) as they find the effect of the ECBs QE-program reached its full effect after 5 quarters. It is therefore possible that the lags in the model are too few.
The more notable difference is the positive effect from the Stibor rate, which is also unexpected when compared with previous studies. This finding does not support the interest rate transmission channel theory. A possible explanation is that the observed dataset contains both the financial crisis which began in 2007 and the euro crisis that followed shortly, and that these have major impacts on the debt levels of Swedish firms that depend on factors outside the Swedish interest rate. This also connects to the findings of Halling, Yu and Zechner (2016) on business cycle dynamics and leverage. If financially constrained firms target leverage ratio evolves differently dependent on their financial constraint, the effect of the interest rate on leverage will in turn depend on their individual financial constraint which is not accounted for in the model. This is part of what could explain the positive effect. Leverage is also measured as a single variable, and the findings by Titman and Wessels (1988) that the type of debt is highly relevant might suggest that debt compositions are needed to explain the results.
27
With the findings of Kühnhausen & Stieber (2014) in mind it is also possible that the model does not encompass all factors that influence firms leverage, as they found evidence of firm size and industry specific growth having a strong impact as well. As such, it is possible that some of the findings, such as the very strong positive effect from the Stibor rate, are due to model misspecifications.
Similar to the leverage regression, the capital expenditure regression showed a negative effect from changes in the Riksbank’s assets and a positive effect from the Stibor rate. Just as for leverage these results are not in line with those produced in previous studies. The R2 is the most notable of the results, however. With an R2 of only 0.74% for the overall sample the regression explains essentially none of the variation in capital expenditure. This is a stark contrast to the 72% reported by Cohen et al. (2019) who used the same model for EMU firm data. It is possible that Swedish firms have other factors that influence their capital expenditure which are needed in the model to explain their investment decisions. The lack of significance for inflation, the high-low spread and the WACC might be further hints at such a possibility as it could indicate that some other factor is distorting their effect. Based on the findings of Blundell-Wignall and Roulet (2013) a measure of firm specific uncertainty might be needed.
While the effect of changes in the Riksbank’s assets were inconclusive for shareholder yield, the finding that a reduction in the Stibor rate had a positive impact was in line with the results for Cohen et al. (2019). A suggestion from these results is that lowering the interest rate has an influence on firm’s shareholder distribution, and that this effect is prevalent despite the lowered cost of investment that follows with it. Therefore, it is possible that the monetary policy carried out by the Riksbank can have an income distributive effect. The main intention of the stimulus is to increase economic activity through lending and investments. As capital ownership is not evenly distributed amongst the population, the subgroup of capital owners will benefit directly if these policy actions result in increased dividends or share buybacks. With the proposed idea of Cohen et al. (2019) that a firm will split its resources into investment and shareholder distribution until the discounted marginal return on a project equals the added value a dividend or buyback has on the firms stock price, it might imply that expansive monetary policy in an environment with few positive NPV investments will benefit capital owners unevenly. The low R2 in the results do indicate however that there might be more factors needed to explain
28
8.2 Limitations and Validity
The findings of this thesis are overall inconclusive as the results are not in line with previous studies and proposed theories. A factor that is important to consider when comparing these results to similar empirical studies is that multiple imputation is not utilized. Multiple imputation is a stochastic estimation method that can be used to fill in missing observations to prevent listwise deletion. As this method is not used in this study, observations that have a missing observation are deleted entirely which creates an unbalanced panel data set. This can have a major impact on the regression. Since multiple imputation can in turn affect the validity of the tests, as it reduces variance and standard errors which can result in positive statistical significance, the choice has been made to not utilize the method. It is possible that part of the difference in results is explained by this choice.
This study is also carried out using self-reported data from Bloomberg, which makes it similar to survey data. An issue with this type of data is the prevalence of data errors due to misreporting. Observing the descriptive statistics in table 1, under section 3, there are possible reasons to suspect prevalence of these data errors. Despite this data having the top and bottom percentiles of observations removed for each variable, variables such as EBITDA and EPS still show extreme values, with a minimum of -2399 for EBITDA and a maximum of 1566.7 for EPS. These are highly unlikely to be realistic observations and it is a possible explanation for their lack of statistical significance in the leverage and shareholder yield regressions. Furthermore, it is possible that their large deviations impact the overall equation. As was shown in the robustness analysis, removing further observations served to reduce the R2 while not showing increased significance for any variables. It is possible however that other data errors that are not abnormal values and therefore unidentifiable are present in the data, which could serve to create noise in the regressions.
29
9. Conclusion
Problems with slow growth and low inflation has prompted several central banks in developed countries to utilize the unconventional monetary policy method of quantitative easing. With its complex transmissions channels and implementation being only just over a decade ago, several of its effects are still not completely understood. This thesis set out to explore the effect that the Riksbank’s monetary policy has had on Swedish firm’s capital allocation decisions by using firm-level data from publicly traded firms. More specifically the aim was to examine if the actions of the Riksbank has encouraged firms to raise their debt level, expand their investments and increase their shareholder distribution.
The findings are somewhat inconclusive and do not fully align with previous research. Leverage and capital expenditure are shown to have a negative effect from increases in the change of the Riksbank’s assets and a positive effect from increases in the Stibor rate. These are opposite to results from previous research on the subject. Shareholder yield shows no conclusive answer on the impact from changes in assets, but a positive effect from reductions in the Stibor rate. This might indicate that parts of the effect of expansive monetary policy are used not for investments but for shareholder distribution instead. As with all empirical studies, the results should be interpreted cautiously. The explanatory power of the regressions are overall low, and there are indicators that more research is needed to understand Swedish firm’s capital allocation decisions in order to fully understand the effects of unconventional monetary policy.
30
10. References
Baumeister, C., & Benati, L. (2013). Unconventional monetary policy and the great recession: Estimating the macroeconomic effects of a spread compression at the zero lower bound.
International Journal of Central Banking, 9(2), 165–212.
Bernhardt, D., Douglas, A., & Robertson, F. (2005). Testing dividend signaling models.
Journal of Empirical Finance, 12(1), 77–98.
Black, F., & Scholes, M. (1974). The effects of dividend yield and dividend policy on common stock prices and returns. Journal of Financial Economics, 1(1), 1–22.
Blundell-Wignall, A., & Roulet, C. (2013). Long-term investment, the cost of capital and the dividend and buyback puzzle. OECD Journal: Financial Market Trends, 2013(1), 39-52. Cohen, A. (1983). Comparing regression coefficients across subsamples: A study of the
statistical test. Sociological Methods & Research, 12(1), 77-94.
Cohen, L., Gómez-Puig, M., & Sosvilla-Rivero, S. (2019). Has the ECB’s monetary policy prompted companies to invest, or pay dividends? Applied Economics, 51(45), 4920–4938. Coibion, O., Gorodnichenko, Y., Kueng, L., & Silvia, J. (2017). Innocent Bystanders?
Monetary policy and inequality. Journal of Monetary Economics, 88, 70-89.
Darracq-Paries, M., & De Santis, R. A. (2015). A non-standard monetary policy shock: The ECB’s 3-year LTROs and the shift in credit supply. Journal of International Money and
Finance, 54, 1-34.
Denis, D. J., & Osobov, I. (2008). Why do firms pay dividends? International evidence on the determinants of dividend policy. Journal of Financial Economics, 89(1), 62-82
Fairchild, R. J. (2006). When Do Share Repurchases Increase Shareholder Wealth? Journal of
Applied Finance, 16(1), 31–36.
Fama, E. F., & French, K. R. (2001). Disappearing dividends: changing firm characteristics or lower propensity to pay?. Journal of Financial Economics, 60(1), 3-43.
Fama, E. F., & Gibbons, M. R. (1982). Inflation, real returns and capital investment. Journal
