• No results found

The impact of financialisation on functional income distribution : evidence from firm-level analysis in Indonesia

N/A
N/A
Protected

Academic year: 2021

Share "The impact of financialisation on functional income distribution : evidence from firm-level analysis in Indonesia"

Copied!
35
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Bruno Besek

MASTER THESIS

The Impact of Financialisation on Functional Income

Distribution:

Evidence from Firm-level Analysis in Indonesia

Track: MSc Development Economics

Course No.: 6414M0228Y

Student No.: 11084529

Supervisor: Prof. dr. Chris T.M. Elbers

2nd reader: dr. Adam S. Booij

(2)

Statement of Originality

This document is written by Bruno Besek who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

SUMMARY

The parallel processes of rising inequality and increasingly developed financial markets in both developed and developing economies raises a question whether these two processes are connected. Using overhead costs, defined as a sum of dividend and interest payments, as a proxy for financialisation, this thesis aims to provide an empirical assessment of the relations between variables forming the Kaleckian theory of mark-up pricing. More specifically, it addresses the possible impact of an increase in overhead costs on the functional income distribution. The analysis is conducted on Indonesian non-financial publicly traded companies from Q1 2000 to Q4 2014. Using the fixed effects estimator and controlling for various estimation problems, the analysis shows that overhead costs do not provide a robust, i.e. consistent and significant, explanation of the changes in the wage share of income. However, the problems arising form the firm-level panel analysis might drive the results.

Keywords: financialisation, functional income distribution, firm-level analysis, panel data, Kaleckian theory of mark-up pricing

(4)

CONTENT

SUMMARY ... III

1 INTRODUCTION ... 1

2 LITERATURE REVIEW ... 3

3 THEORY ... 5

3.1 Financial vs physical investment ... 7

3.2 Determinants of the price mark-up ... 8

3.3 Predicted effect of variables ... 9

4 DATA ... 12

5 ESTIMATION STRATEGY ... 13

6 DESCRIPTIVE STATISTICS ... 16

7 RESULTS ... 18

8 DISCUSSION ... 23

8.1 Theory and real-life data ... 24

8.2 The appropriateness of the econometric method used ... 24

8.3 The structure of variables ... 25

9 CONCLUSION ... 27

10 REFERENCES ... 29

11 LIST OF TABLES ... 30

12 LIST OF FIGURES ... 30

(5)

1

1 INTRODUCTION

The process of financialisation in developed economies started at the beginning of the 1980s. It came as a result of broad liberalisation measures in the goods’, labour and financial markets as well as a result of lowering barriers to trade. Epstein (2005) defines financialisation as „the increasing role of financial motives, financial markets, financial actors and financial institutions in the operation of domestic and international economies“ (Epstein, 2005, p3). In the same time an increase in inequality and slowdown in growth have been encountered as well. These observations have led economists to develop a number of stylized facts about financialisation which can be observed as (Hein, 2015):

1. a rise in gross profit share (including retained profits, dividends and interest payments) 2. an increase in shareholder power vis-a-vis firms and workers1

 because of the higher relative mobility of capital, shareholders can change their place of business easier than firms and workers

3. an increase in the rate of return on equity and debt held by rentiers

 this is the result of the first stylized fact but also because ever increasing amounts of debt tend to increase the rate of return on bonds due to higher risk associated with the higher amount of debt

4. an alignment of management’s with shareholder’s interests

 the emergence of new forms of managers’ remuneration (e.g. stock options) 5. an increasing potential for wealth-based and debt-financed consumption

 the growth in the value of shares increases the probability of shareholder getting a loan (which can be used for consumption)2. In the same time, in order to keep the desired level of growth firm increases its level of debt3.

These developments and their impact on inequality and growth have been mostly studied in developed economies4. Thus, it is still uncertain whether they have spilled over to the developing world. Some authors and institutions suggest that the impact is significant in

1

Where Hein presumably refers to shareholders as external forces influencing a firm such as e.g. banks and in the same time interprets firms as establishments run by managers, thereby making a distinction between an establishment and workers.

2 One of the basic models of the pricing of stocks, Gordon’s model, suggests that the price of the stock is

positively correlated with the dividend paid (Brigham and Houston, 2004). An increase in the price of stocks makes a stockholder more credit-worthy because the value of assets at her disposal increased.

3 As explained in the third stylised fact the increase in the level of debt may increase the interest payments.

Thus, firms are believed to be in search of ever higher profit margins to satisfy growth levels, pay-out ratios and increased interest payments. The relationship is further explained in the Theory.

4

(6)

2

developing world as well. UNCTAD Trade and Development Report (2015) proxies financialisation with capital flows to developing and emerging economies and concludes that „capital inflows exert pressures for real exchange rate appreciation and elevate the primacy of short-term returns in speculative markets over long-term projects that raise productive capacity“ (UNCTAD, 2015, p39). ILO (2011) asserts that the decline in labor share of income since the 1990s is more pronounced in developing and emerging economies than in developed economies. In the same period, however, emerging economies have already developed their capital markets tremendously5 (Schellhase, Sau and Prabha (2014)) which brings upon the parallel processes of rapidly developing financial markets and rising inequality mentioned above and raises the question of whether they are connected.

However, as the literature review below suggests, there is no study assessing the impact of financialisation on inequality in developing economies using firm-level microeconomic data. Therefore, the analysis below tries to provide an answer to the research question of what is the impact of financialisation on functional income distribution in Indonesian non-financial publicly traded companies. In the spirit of the Kalecki’s theory of mark-up pricing, overhead costs will serve as a proxy for financialisation, referring to financialisation as a rise in gross profit share (first stylized fact). The purpose of the analysis is to check whether financialisation influences functional income distribution in Indonesia and to give a further stimulus in establishing financialisation as one of the explanatory variables of the wage share of income6.

The thesis is structured as follows. The current research on the impact of financialisation on functional income distribution is presented in section 2. Section 3 focuses on theories underlying the estimation strategy. The data is decribed in the section 4. The estimation strategy is thoroughly explained in the section 5 along with problems that can be encountered when analysing firm-level panel data. After that, descriptive statistics and results are presented in section 6 and 7, respectively, which is followed by the discussion and conclusions drawn from the analysis.

5

Few countries have more than 400 domestic companies listed on the corresponding stock exchanges (China and India are outliers with 2,613 and 5,541 listed companies respectively). Some countries even have a market capitalization greater than their GDP (e.g. South Africa, Thailand, Malaysia)

6

The term wage share of income is used because the labour share of income is a broader term which is adjusted for the income of self-employed and which could not be analysed with the data used.

(7)

3

2 LITERATURE REVIEW

A number of studies examine the relationship between finance and inequality. According to Demirguc-Kunt and Levine (2009) financial development works on intensive and extensive margin (“Financial development might operate both on the intensive margin by improving the financial services employed by those already accessing the financial system. Finance might also influence the extensive margin by allowing poorer families to reduce liquidity constraints, expand investment opportunities, and manage risk.“ (Demirguc-Kunt and Levine, 2009, p21)). Thus, according to them, financial development can simultaneously help the poorest families to escape the vicious circle of poverty but it also allows wealthy individuals gain even more wealth.

They further provide a survey of the corresponding literature and conclude that improvements in financial contracts, markets, and intermediaries usually reduce inequality. Acknowledging this point of view, the other side of the story, namely the relationship between financial development and the rising wealth of the upper parts of income distribution, has been given little attention in mainstream journals. Nevertheless, studies analysing the effect of financialistion on either labour market institutions or labour share of income have been undertaken by a couple of economists belonging to the post-Keynesian school of thought. For example, using a country fixed effects estimator on a sample of 16 OECD countries in the period from 1970 until 2009 Darcillon (2015) argues that an increase in employment and value added of financial intermediation industry7 reduced workers’ bargaining power and lessened the strength of employment protection legislation. Dünhaupt (2013b) also uses a fixed effects estimator on a panel of 13 OECD countries in the period from 1986 to 2007 to study the effect of financialisation on labour share of income. In her study she uses a number of proxies to control for the effect of globalisation and labour bargaining power on adjusted labour share. In the end she concludes that financialisation (measured as net dividend and interest payments of non-financial corporations) has a negative effect on adjusted labor share. In a similar study, Dünhaupt (2013a) provides an overview of how different economic theories approach the analysis of functional income distribution and presents the current research on changes in it. Once again, she concludes that financialisation should be taken into account when analysing the functional income distribution. Following up on that, Alvarez (2015) argues that a 1% increase in financialisation, measured by company’s

7

(8)

4

financial profits8, reduces the wage share of income by 0,4%. To prove this, he uses microeconometric fim-level panel data on French non-financial publicly traded companies from 2004 to 2013. Along with significant and negative impact of financialisation on wage share of income, he also finds that proxies for globalisation (export revenues), technological change (fixed asset value) and labour bargaining power (firm-level employment) affect wage share of income in a significant and expected way9. Returning to the country-level analysis, Stockhammer (2012) studies the effect of technological change, globalisation, financial globalisation and welfare state retrenchment on functional income distribution. He uses a panel of 71 developed and developing countries from 1970 to 2007 and, like all three authors mentioned above, applies the fixed effects estimator in his analysis. He concludes that among the effects of four variables studied, financial globalisation has the strongest effect on the adjusted wage share. In one of his specifications he uses 5-year averaged data, a feature similar to the one used in the analysis below.

As evidence that financialisation has spilled over to emerging markets, Demir (2009) finds that non-financial corporations in Argentina, Mexico and Turkey have already started switching their corporate strategies from physical to financial investment. Using firm-level panel data, he argues that the increase in the rate of return on financial investment and the overall uncertainty and risk have a negative effect on physical investments. He also introduces two new stylized facts of financialisation which are more closely related to the firm-level. These are:

6. an increase in the acquisition of short-term financial assets by firms from non-financial sectors

7. a decrease in physical investment rates

Studies mentioned above do not suggest that financialisation is the sole explanatory factor of functional income distribution. Jaumotte and Tytell (2007) studied how labour globalisation in terms of trade prices, offshoring and imigration affected labour share in advanced economies. They also used a number of variables to control for the effect of technological change and changes in labour market policies on labour share. In the end they concluded that labour globalisation had a negative effect on labour share in advanced economies, but not in such a great extent as technological change affected it. Ferreira, Leite and Wai-Poi (2007) used household data from Brazil during the period of trade liberalisation (1988-1995) to

8 Financial profits are profits accruing from dividends, interest payments and capital gains. 9

Alvarez (2015) finds that export revenues and fixed assets have a negative, while firm-level employment has a positive effect on wage share of income.

(9)

5

study the effect that trade liberalisation measures had on wages and employment. They did so by employing a two-stage analysis where in the first stage they estimated the wage premia and the probability of a person being in a specific occupation, whereas in the second stage they used the coefficients estimated in the first stage to study the effect of trade liberalisation measures on wage premia and employment. Their analysis brought them to the conclusion that trade liberalisation measures have affected inequality mainly through changes that trade liberalisation measures had on employment across sectors. Furthermore, studying the determinants of functional income distribution in Swiss firms, Siegenthaler and Stucki (2014) assert that “functional income distribution is ultimately determined at the firm level”10

. Much as like Darcillon (2015) argues that changes in the employment and value added of financial intermediation industry bring changes in labour market institutions, they further argue that changes in the composition of the whole industry/economy affect its labour share of income. Therefore, the aim of this study is not to single financialisation out as the main determinant of functional income distribution but to continue on the work started by Alvarez (2015), Dünhaupt (2013a, 2013b), Demir (2009), Stockhammer (2012) and Darcillon (2015) in establishing financialisation as one of its determinants.11

As discussed studies suggest, financialisation can be proxied by numerous variables. Particularly, it can be proxied by any variable mentioned in the seven stylized facts of financialisation. The analysis below focuses on the first stylised fact and interprets financialisation as a rise in gross profit share (measured as a sum of dividend and interest payments).

3 THEORY

Theoretical papers studying the impact of financialisation on output growth and labor share of income usually use the post-Keynesian theory of firm as their starting point. This approach is different from the conventional approach as it recognizes the internal power struggle as the central point in the firm analysis instead of seeing a firm as a profit maximizing tool. The internal power struggle rests on different interests that various groups within a firm have. By assumption, shareholders have a short-term perspective on firm’s profitability, while workers and managers have a long-term perspective on firm’s growth and survival (Stockhammer

10

One of the main researchers of functional income distribution and inequality, Anthony Atkinson, asserts that the discussion about the functional income distribution is as important today as it was when there existed a clear distinction between the receivers of rents, profits and wages. (Atkinson, 2009)

11

As Alvarez (2015) puts it: “to explain the differences in wage shares at the firm level one must take into account other dimensions ..., particularly those related to the financial sphere.” (Alvarez, 2015, p470)

(10)

6

(2004, 2006)). Even without using this assumption, Dallery (2008) manages to obtain the same outcomes as when the assumption holds. He introduces a new constraint in firm’s internal funding because now managers need to satisfy shareholders’ interests by distributing more dividends (shareholder value orientation). After that, he presents even more strict approach which not only acknowledges the existence of shareholders as the new constraint in firm’s internal funding but also suggests that these same shareholders exert pressure for the profit-maximization strategy in order to have a greater amount of profits to demand dividends from. This way they inhibit the firm from its long-run growth potential, which is explained by the analysis presented in Dallery (2008, p5). He starts off by firm’s dilemma of how to allocate its scarce resources:

iD

x Isx Id  I x If  (1 sf)(

iD) ,

where I is physical investment, π profit before interest payments, i interest rate, D stock of debt, sf retention rate12, xs, xd and xf new net share issues, new net debt and financial

investment13, respectively all expressed as a ratio of physical investment. This identity shows that if a firm increases its financial investment or decreases its retention ratio, they need to either increase their profits, raise their equity or debt level14, or decrease their physical investment. From this identity Dallery (2008) further explains how to relate firm’s profit rate with its capital stock growth rate and concludes: “if managers want to grow at the same rate, and if they are obliged to distribute more dividends than before, they have to reach a higher profit rate (margin) to finance their investment projects“ (Dallery, 2008, p11) (Figure 1). In Figure 1 the finance frontier shows the accumulation rate (g) of capital stock that a firm can achieve with a certain profit rate (r) if it reinvested all of its profits. Expansion frontier, on the other hand, shows what is the expected profit rate if the capital stock of a firm grows at a certain accumulation rate. According to studies done by Stockhammer (2004, 2006) and Dallery (2008), managers prefer higher accumulation rate, while shareholders prefer higher profit rate. Thus, depending on the respective bargaining power, the end point will usually be somewhere in between “Manager’s preference” and “Shareholders’ preference” points.15

The expansion frontier allows for an interesting analysis because it juxtaposes micro- and macroeconomic approach. If a firm cuts back its investment it moves along the expansion

12

Retention rate measures the proportion of profits that stays in the firm as retained profits.

13 Financial investment comprises of investments in financial instruments, such as shares, bonds, options, etc. 14 An increase in equity hurts existing shareholders, while an increase in debt makes a firm riskier.

15

Firms adopting a shareholder value strategy face an upward shift and a counter-clockwise rotation of the finance frontier because shareholders capture part of the profit rate.

(11)

7

frontier. But, if firms do so at a larger scale aggregate demand will fall and profit rate will decrease at each rate of accumulation. Similarly, a firm that wants to reach a higher profit level can, e.g., cut wages. But, at macroeconomic level, if many firms cut their wages aggregate demand will fall and so will the profit rate for each level of accumulation (Dallery and van Treeck, 2009). Referring to Keynes: “We have here an extreme example of the disharmony of general and particular interest...” (Keynes, 1932).

Figure 1 The Post-Keynesian firm and the shareholder-manager conflict

Source: Dallery and van Treeck (2009, p5)

3.1 Financial vs physical investment

Since 1980s financial investment has become a substitute for physical investment. Firms started to switch their strategies towards financial investment because physical investment usually brings more remote rates of return than financial investments and because of the inherent uncertainty and lock-in effect of physical investment. Demir (2009) found a proof of this switch for Argentinian, Mexican and Turkish firms. Financial investment in these countries is even more attractive because of the high government debt which usually leads to higher interest burden. Although Tobin (1965) stated that “financial assets ultimately relate to the productive investments of the companies”, “the seven market failures in financial markets” discussed in Stiglitz (1993) inhibit financial investment to be a mere catalyst for physical investment. Thus, financial investment is not only a substitute for physical investment but it can reduce growth and accumulation, as presented by the graph above. Demir (2009) also mentions credit bottlenecks and profitability squeeze as explanations for the strategy switch.

(12)

8

Shareholder’s short-termism puts pressure on managers to develop more short-term objectives, to adopt short-term strategies and correspondingly to switch their strategy towards in short-term more profitable investment (which has been financial investment since the 1980s).

3.2 Determinants of the price mark -up

In his study of financialisation in France Alvarez (2015) argues that the connection between the switch in firm’s strategies, as a feature of financialisation, and wage share of income goes through the price mark-up.

According to Kalecki (1954) and adapted by Hein (2015), this mark-up depends on:

1. the degree of price competition in the goods market

 price competition decreases as the degree of market concentration increases (Kalecki, 1954, p17)16. Taking marginal costs as given, the inability of the firm to change the price of its product makes increasing shareholder pressure to squeeze out other determinants of the mark-up.

2. the bargaining power and activity of trade unions 3. overhead costs and gross profit targets

 overhead costs include dividend and interest payments 4. the price of imported raw materials and semi-finished products 5. the sectoral composition of domestic economy

By this mark-up decomposition, it is evident, as Hein (2015) puts it, that “functional income distribution is thus determined by the mark-up in pricing of firms, by the relationship of unit material costs to unit labour costs and by the sectoral composition of the economy” (Hein, 2015, p923).

Combining the analysis in Dallery (2008) with mark-up determinants, it seems reasonable to argue that an increase in overhead costs would have a negative effect on wages when prices are constant. In theory, constant prices should be expected because deviating from the equilibrium price and quantity leads to lower profits. Other determinants of price mark-up,

16 It can also be argued that even when the degree of market concentration decreases, price competition may

decrease as well, as the institutional setting resembles that of monopolistic competition. This setting allows both, the degree of concentration and price competition to be low as a lot of firms do not compete with prices, but rather with product differentiation, marketing, advertising, etc.

(13)

9

namely sectoral composition and the price of imports, are considered exogenous at firm level17.

3.3 Predicted effect of variables

3.3.1 The Kalecki’s theory of price mark-up

The relationship between overhead costs and wage share of income will be studied by applying the Kalecki’s theory of mark-up pricing. Kalecki’s representation of the theory resembles an identity in which, holding everything else equal, it is evident that an increase in overhead costs (mark-up) will lead to a decrease in wages18. Interested in the extent of the relationship between overhead costs and wages, and bearing in mind that an identity derived from a theory, does not need to hold in practice, I will start from a general specification of Kalecki’s theory:

, ,

pf w m

where p is the price per unit of output, µ is the mark-up over marginal cost pricing reflecting a number of things mentioned previously, w is the wage per unit of output and m represents material costs per unit of output. 19

It follows that the wage function is:

( , , ) wg p m , which, upon the application of natural logarithm becomes:

0 1 2 3

lnw  lnp lnm ln

. (1)

This leads to the specific form of Kalecki’s theory:

3 1

1 2 1 /

1/ /

pCwm   , (2)

17 Sectoral composition and the price of imports are controlled for by firm and time fixed effects since imported

raw materials are usually of the same price for each firm, while the sectoral composition of the economy does not change quite often. The use of firm and time fixed effects is further discussed in Results.

18 Kalecki's first specification of the relationship was pm u n p** , where u represents prime costs

(material costs and wages) per unit of output, p represents the average price per unit of output of all firms in an industry, while m and n are coefficients (Kalecki, 1954).

Further analyses of Kalecki's theory bring specifications such as one used by Hein (2015):

(1 )( )

j j j j j

p  m wap e ,

which, in a simplified version though, is the specific form of the theory this study will deal with. In this equation, aj is the inverse of the labour productivity, pj the price of the imported raw materials and semi-finished

inputs in the foreign currency, e the exchange rate and µj imported raw materials and semi-finished inputs per

unit of output.

19

(14)

10

where C is a constant. According to the theory, an increase in each of the RHS variables should lead to an increase in price. This poses conditions upon the coefficients, but also suggests the predicted effect of each variable on wages:

β1 > 0

β2, β3 < 0,

Referring to equation (1) above, an increase in unit overhead costs (mark-up) is predicted to lead to a decrease in unit wage.

3.3.2 The Solow growth model

The Solow growth model provides an interesting framework for studying the functional income distribution. By relaxing the assumption of exogenously set wages(w) and rents(r), the main variables from the Solow model (savings, depreciation rate and capital-labour ratio) can be used as determinants of functional income distribution.

Figure 2 makes it clear that an increase in savings will have an ambiguous effect on the distribution of income (y=rk+w). The similar story applies to depreciation rate(δ) and capital-labour ratio(k). E.g. an increase capital-capital-labour ratio increases the share of capital just because there is more capital divided among the same number of people. On the other hand, the neoclassical model suggests that at the same time when capital-labour ratio increases the return on capital should decrease. Thus, the final effect is ambiguous.

Source: Lectures in the Growth Models course at the Faculty of Economics and Business, University of Zagreb held by prof.Ivo Bićanić, February-June 2015 Figure 2 Functional income distribution explained by the Solow growth model

(15)

11

3.3.3 Variables suggested by the literature

The Solow model brings me to the variable which, according to the literature, influences functional income distribution as well. Technological change can alter the distribution of income because it can be either capital or labour enhancing. It is usually proxied by research and development (R&D) costs but when this proxy is applied in the analysis below the sample size drops considerably. Therefore, capital expenditure is used instead. It is assumed that when firms acquire new capital, they acquire the latest technology that is available. There are three reasons why this might be highly questionable. First, firms may lack resources to buy the latest technology accessible, so they will often buy second-hand technology. Second, capital expenditure in the dataset may include acquisitions of other firms, which may use either new or old technology, so capital expenditure may not represent technological change as well as the R&D cost or the use of ICT capital20. Finally, according to the Solow model capital-labour ratio, savings and depreciation rate are all determinants of a change in capital stock, which is closely related to capital expenditure. In the same time, capital expenditure has a direct effect on the level of capital. Therefore, a multicollinearity check is needed to make sure that all variables drawn from the Solow model are precisely estimated.21 But, as the main focus of my analysis does not lie in the estimation of coefficients of these variables, multicollinearity between them should not be a major cause for concern if it cannot be rejected.

Literature suggests that globalisation influences functional income distribution as well. However, its impact on functional income distribution is ambiguous in the literature. The neoclassical Stolper-Samuelson (SS) theorem suggests that opening up of the economy will benefit the factor which the economy is abundant in (Stolper and Samuelson, 1941). So, in the context of a developing economy, the more open an economy is, the larger are the benefits for labour because developing economies are assumed to be abundant in labour. However, Stockhammer (2012) finds that the negative effect of globalisation on wage share can be found in both, developed and developing economies. Alternatively, the Political Economy approach to trade, used by Rodrik (1997) and Onaran (2011), suggests that a relatively more mobile factor has greater bargaining power, and thus benefits from

20 The dataset did not allow for the decomposition of capital expenditure.

21 „... when multiple regressors are imperfectly multicollinear, the coefficients on one or more of these

regressors will be imprecisely estimated - that is, they will have a large sampling variance.“ (Stock and Watson, 2011, p202)

(16)

12

globalisation more. Since capital is more mobile than labour, the wage share of a country is expected to fall as the country increases its exposure to international trade.

Applying the SS theorem to firm level, it can be argued that a firm using relatively more labour in its production will see its wage share to increase as it gets more involved in the world trade (e.g. has more exports22). International sales will serve as the proxy for the firm’s openess. One specification will also include the variable interacting export revenue and capital-labour ratio assuming that when company engages in the world trade its wage share of income will rise more when its capital-labour ratio is lower.

Since wage share is countercyclical, GDP growth rate should be included to control for business cycles. These are time-period specific, i.e. they are common to all firms in each period. (Stockhammer, 2012)

4 DATA

The major issue with the data was to find a developing country with enough publicly traded companies that report data on wages. Since this study was based on a firm-level analysis, a minor issue was to find a country where financialisation is already in progress. The macroeoconomic implications can afterwards be inferred from the microeconomic analysis because as Siegenthaler & Stucki (2014) put it: “... ‘functional income distribution’ is ultimately decided within individual firms.” (Siegenthaler and Stucki, 2014, p3)

Even though Demir (2009) concluded that companies from Argentina, Turkey and Mexico have already adopted financialisation-related strategies, companies from these countries lack data on wages and employees23, especially for the period studied by Demir. Therefore, this study analyses 409 Indonesian firms that reported the data on wages (402 on employees) from Q1 2000 until Q4 2014. It is important to mention that the sample is highly unbalanced because firms need to start publishing their data only from the period they go public and they did not go public at the same time24. After deleting all periods before a firm went public (observations with missing values in all variables within a firm), the sample reduced to 19,652 observations.

The data was gathered from the Compustat Wharton WRDS database, and was supplemented by data from the Worldscope and the Datastream databases if certain variables lacked in the

22 This cannot be said for the ratio of imported raw materials and semi-finished products, however. According to

section 3.2. an increase in imported inputs creates a downward pressure on wages holding everything else constant.

23 There are 68 Argentinian firms with data on wages (40 with data on employees), 58 Mexican firms (87) and

100 Turkish firms (101).

24

(17)

13

former. Companies were merged by their International Securities Identification Number. The analysis was performed on publicly traded non-financial companies from Indonesia since their main focus of business is not financial. Thus, their increased dependence on financial instruments presents a proxy for financialization. The focus on one country is to avoid any differences between economic systems of different countries.

Different volatility of data between variables poses a problem in the analysis. E.g. data on wages, employment, finished goods, materials and international sales varies on a yearly basis, while data on dividends, interest, depreciation, total assets, retained earnings, sales and capital expenditure varies on a quarterly basis. Due to this reason, the analysis uses four period lagged, equally weighted moving averages of those variables that vary on a quarterly basis. The current observation is not included because it is assumed that wages cannot be set in the same period when a firm observes its financial results.

Also, since a minor number of firms published their financial results in US dollars, these results were converted to Indonesian rupiahs using quarterly averaged daily exchange rates between the two currencies.25 In order to make financial results comparable across the observed period, all data is expressed in 2010 Indonesian Rupiahs as well.

5 ESTIMATION STRATEGY

With the insight from theory, the main estimation specification is:

, 0 1 , 2 , 3 ,

lnwi t   ln(overheads)i t lnXi ts  lngdpt  i i t, (3)

where wi,t represents wages per unit of output in company i in period t, Xi,ts represents the

vector of covariates26, gdpt represents the GDP growth rate in year t, λi represents firm fixed

effects and εi,t represents the estimation error. Firm fixed effects are included to control for

any unobserved variation between firms that arises because of firm-specific time-invariant characteristics.

Figure 3 shows the development of the logarithm of unit wage and the logarithm of unit overhead costs among Indonesian non-financial publicly traded companies. Since presenting the time series for each company would make graphs illegible, both variables are averaged across all companies in a specific quarter. Thus, the left graph shows the unweighted average

25

The data on daily exchange rates was dowloaded from http://www.imf.org/external/np/fin/data/param_rms_mth.aspx

26 Vector X

i,ts consists of unit sales, unit material costs, capital-labour ratio, depreciation rate, retained earnings,

unit international sales and capital expenditure. Unit sales are sales per unit of output and thus represent a proxy for price.

(18)

14

of the logarithm of unit wage and the right graph presents the unweighted average of the logarithm of unit overhead costs.

Graphs in Figure 3 clearly present the problem of different volatility between variables discussed above. More frequent changes are spotted in the right graph where overhead costs change on a quarterly basis.2728 Moving averages will help in controling for differences in volatility but one specification including an interaction variable between overhead costs and quarter dummies will also be applied to control for the seasonality of dividends. Another observable pattern from the graphs above is a downward trend in both variables, suggesting that these two variables might not move in the direction predicted by equation (1). However, in order to make this kind of a conclusion, a more thorough analysis than simple eyeballing of the data should be applied.

A closer look at the data reveals that the problem of non-stationarity in the logarithm of unit wage might arise. In order to test for unit root, a Fisher-type panel unit root test was

27 Figure 3 shows variables in constant IDR but different volatility between two variables is even more easily

observed when using non-constant values (not shown). E.g. flat behaviour of wages during the year is easily observed on a graph showing non-constant values. The logarithm of unit wage varies on a quarterly basis because of quarterly changes in output.

28

This represents a common pattern in the financial analysis because dividends are usually paid out in the second quarter.

Source: Compustat - Capital IQ, Datastream

Figure 3 Time-series of the logarithm of unit wage (left graph) and the logarithm of unit overhead costs (right graph)

(19)

15

applied29. Its null hypothesis states that all panels are non-stationary, while the alternative hypothesis states that at least one panel is stationary (Baltagi, 2005). The null hypothesis can only be rejected when the drift parameter is included, suggesting that the logarithm of unit wage is stationary in at least one panel only when the constant is included in the first-order autoregressive equation.30 However, the disadvantage of the test is that the number of time periods should be much bigger than the number of entities in the panel which is not the case in this study. But, as the Fisher-type unit root test is the only panel unit root test allowing the panel to be unbalanced, I decided to use it in the analysis. On the other hand, when testing for unit root in the logarithm of unit overhead costs the null hypothesis is rejected even without using the drift parameter in the first-order autoregressive equation. Although both variables, when averaged out by time period, show trend behaviour, the Fisher-type unit root test does not suggest that trend is a problem when analysing separate panels. In order to correct for non-stationarity in the logarithm of unit wage, the following specification will be applied:

, 0 1 , 2 , 3 ,

lnwi t   ln(overheads)i t  lnXi ts  lngdpt  i i t

       . (4)

The only difference between this specification and specification (3) is that the difference of the logarithm of unit wage is used as dependent variable.

Another way to approach the problem of non-stationarity is to apply different type of standard errors, namely Driscoll-Kraay standard errors. These standard errors control for heteroskedasticity, autocorrelation and cross-sectional dependence (Alvarez, 2015, p466) and thus, could revert the analysis back to specification (3) which is more consistent with the theory. The Fisher-type unit root test assumes data to be cross-sectionally independent, an assumption that is highly questionable when using firm-level data analysis31. Thus, since Driscoll-Kraay standard errors control for cross-sectional dependence, they are more appropriate than cluster robust standard errors that do not control for it. In any specification that does not apply Driscoll-Kraay standard errors, standard errors are clustered at firm level.

29 „Fisher-type test [...] combines the p-values from unit root tests for each cross-section i to test for unit root in

panel data.“ It „can be applied to any other unit root tests.“ (Baltagi, 2005, p244)

30

This means that the first difference of the logarithm of unit wage will follow a random-walk process around some average and not around zero.

31 In the firm-level analysis independence across firms (panels) cannot be assumed because one firm's inputs are

another firm's output, so firms are essentially forming a network of connections. The data also did not allow me to perform Pesaran test for cross-sectional dependence.

(20)

16

6 DESCRIPTIVE STATISTICS

Table 1 presents main summary statistics. It can be immediately observed that variables have different number of observations which means that the sample is highly unbalanced32. Not only do the variables contain different number of observations but so do the companies since they were not obliged to report the data during the whole period. As discussed above, they are only obliged to do so when they go public and they go public after they reach a certain threshold in terms of assets, profits or some other indicator. In other words, when deciding whether to go public or not, a firm considers variables that may influence wages as well. This creates a sample selection bias problem as the sample consists of only those firms that are publicly listed on the stock exchange and they most likely financially outperformed those firms that are not publicly listed. The self-selection of firms in the sample might be partially solved by controlling for company’s assets in the period a company went public. Thus, the analysis below includes one specification in which all variables are inversely weighted by assets a company owned when it went public. Alternatively, Dionne, Gagné and Vanasse (1998) propose to jointly estimate both, the probability of company’s entrance/exit and the main function, whatever the main function is.33 Nonetheless, due to the lack of variables predicting firm’s entry to or exit from the stock market, entrances and exits of companies had to be treated as random34.

Table 1 Summary statistics

All variables expressed in 2010 IDR. The natural logarithm applied to all variables. Wages, Overheads, Sales, Materialc & Int_Sales expressed per unit of output

Retained & Capex expressed per Added value (=interest+cost of employees+depreciation+gross profit/loss+tax) Depreciation expressed per unit of capital

Materialc - material costs; CL - Capital/Labour ratio; Depreciation - depreciation rate; Retained - retained earnings; Int_Sales - international sales; Capex - Capital expenditure

32

„In practice, most (if not all) panel data sets of firms are incomplete.“ (Dionne, Gagné and Vanasse, 1998)

33 This method is deferred to further studies because of its complexity.

34The same authors analyse a commonly applied solution by researchers when dealing with unbalanced panel

data, sub-balancing, i.e. excluding firms with incomplete time-series (Dionne, Gagné and Vanasse, 1998, p310). However, this method would reduce the sample greatly.

Logarithm of N mean sd min max

Wages 11066 .3448151 1.524575 -5.149018 9.49238 Overheads 6953 -9.615813 2.159384 -21.90671 .3314828 Sales 7704 -5.42287 1.512026 -14.39267 3.003705 Materialc 5005 2.131966 1.760687 -10.4241 7.246875 CL 13148 5.747246 1.688378 -4.097462 14.68904 Depreciation 9728 -3.656093 1.314117 -13.87687 5.744275 Retained 5803 .885671 1.413539 -7.732123 9.525006 Int_Sales 2684 1.172204 2.320052 -14.23353 11.45477 Capex 7553 -1.226984 1.776443 -11.34685 10.84896 N 16356

(21)

17

Table 1 also clearly demonstrates the large variation within variables. Some of the variation will be controlled for by using the fixed effects estimator but a large part of variation may arise within a company over the period studied. A deeper analysis of some variables showed that a large part of variation within variables comes from outliers, so the specification with variables trimmed at 1st and 99th percentile is performed.

Table 2 Panel variation of the sample

Logarithm of1 Mean Std. Dev. Min Max Observations Wages overall .3448151 1.524575 -5.149018 9.49238 N = 11066 between 1.663281 -3.772914 6.400296 n = 317 within .7398791 -4.322665 4.439729 T-bar = 34.9085 Overheads overall -9.615813 2.159384 -21.90671 .3314828 N = 6953 between 2.036192 -14.35808 -1.775399 n = 301 within 1.395845 -22.99089 -1.129318 T-bar = 23.0997 Sales overall -5.42287 1.512026 -14.39267 3.003705 N = 7704 between 1.46437 -9.382191 -.4910291 n = 303 within .8698247 -12.69381 -.1009608 T-bar = 25.4257 Materialc overall 2.131966 1.760687 -10.4241 7.246875 N = 5005 between 1.492686 -3.955292 6.547275 n = 176 within 1.210754 -9.97939 5.944148 T-bar = 28.4375 CL overall 5.747246 1.688378 -4.097462 14.68904 N = 13148 between 1.378403 2.29597 9.874415 n = 402 within 1.03212 -4.227836 13.13006 T-bar = 32.7065 Depreciation overall -3.656093 1.314117 -13.87687 5.744275 N = 9728 between .802628 -8.387682 -1.082923 n = 378 within 1.100642 -13.08271 5.325577 T-bar = 25.7354 Retained overall .885671 1.413539 -7.732123 9.525006 N = 5803 between 1.403962 -3.07667 7.264536 n = 351 within 1.016971 -7.30009 8.941036 T-bar = 16.5328 Int_Sales overall 1.172204 2.320052 -14.23353 11.45477 N = 2684 between 2.388667 -6.078873 7.947333 n = 153 within 1.186276 -10.82724 8.703985 T-bar = 17.5425 Capex overall -1.226984 1.776443 -11.34685 10.84896 N = 7553 between 1.12448 -6.078873 2.905903 n = 368 within 1.491893 -10.96878 10.02357 T-bar = 20.5245 1

See Table 1 for the explanation of variables.

Table 2 confirms the huge variation between and within variables encountered in Table 1. Since the main estimation specification is in a logarithmic form, all variables are presented in logarithms. It can be immediately observed that most variables used in the analysis have

(22)

18

between variation more pronounced than within variation, meaning that variables do not vary so much through time as they vary across firms. This may pose a problem in the estimation using firm fixed effects because they control for firm specific time invariant characteristics. As the bigger part of variation within most variables is between companies rather then within them, there may be a lack of variation to estimate the coefficients appropriately.

Moving on to Table 3 shows that the econometric analysis below might not predict the effect of variables on unit wage in a way the Kaleckian mark-up theory suggests. Table 3 presents pairwise Pearson correlations between variables, which show that all variables suggested by Kaleckian mark-up theory are positively correlated. But, according to the theory, positive correlation should only be expected between Sales and Wages. However, two things are important to mention. First, these are pairwise correlations that do not control for any other variable in the table, entity or time fixed effects nor the fact that the logarithm of unit wage is non-stationary. Second, the table is presented in order to obtain the initial insight in the relationship between variables and it is not the proof of the absence of multicollinearity between variables.35

Table 3 Pairwise correlation of main variables Logarithm of1 Wages Over- heads Sales Material costs CL ratio Depre- ciation Retained Int_ Sales Capex Wages 1.00 Overheads 0.36*** 1.00 Sales 0.78*** 0.41*** 1.00 Material costs 0.62*** 0.36*** 0.58*** 1.00 CL ratio -0.22*** 0.10* 0.02 -0.01 1.00 Depreciation 0.06 -0.09* -0.05 -0.08 -0.27*** 1.00 Retained 0.03 -0.21*** 0.03 -0.04 0.07 -0.04 1.00 Int_Sales 0.56*** 0.37*** 0.48*** 0.34*** -0.02 -0.03 -0.07 1.00 Capex 0.03 0.16*** 0.12** 0.07 0.28*** -0.06 0.26*** 0.06 1.00 * p < 0.10, ** p < 0.05, *** p < 0.01

1See Table 1 for the explanation of variables.

7 RESULTS

The Kaleckian mark-up pricing theory and the Solow growth model recommend using different variables in explaining functional income distribution, more specifically in explaining the wage share of income. Therefore, it would be interesting to see which theory provides a better explanation of the wage share of income. It may also be the case that both

35 Appendix provides a more formal test of multicollinearity which shows that multicollinearity does not

(23)

19

theories, including variables suggested by the literature (technological change and globalisation), are useful in explaining it.

In order to establish whether there is a causal relationship between variables, Table 4 provides regressions where each regression is enriched with variables from another theory. The regression in the first column uses variables from the Kaleckian theory only, the regression in the second column is enriched with variables from the Solow growth model and the regression in the third column with variables suggested by the literature. It is important to mention that these estimations neither control for non-stationarity of the logarithm of unit wage nor for the different volatility between variables.

As suggested by Table 3, unit overhead costs are positively related to unit wage in Table 4 as well. It is also observed that variables suggested by the Kaleckian theory are significant in almost all cases, except when international sales and capital expenditures are controlled for. However, only unit sales have the expected sign. This is why these initial regressions cannot tell which theory explains functional income distribution better.

Table 2 revealed that most variables vary relatively more between firms, rather than within firms. This is observed in Table 4 as well since the standard deviation of the individual effect (sigma_u) is bigger than the standard deviation of the estimation (idiosyncratic) error (sigma_e) in all three specifications36. Moreover, Dionne, Gagné and Vanasse (1998) state that in the short panel case the use of random firm specific effects and fixed time specific effects is preferred to the use of fixed firm specific effects. Whether firm fixed effects should be used or not is checked by the Hausman robust test37. Its null hypothesis could be rejected at 1% significance level which means that firm fixed effects estimator should be used.

Another reason why firm fixed effects might be more appropriate is that they control for the unobserved, time-invariant variation between companies. The unobserved variation between companies arises due to the lack of proxies for variables specified in the Kaleckian mark-up theory, such as e.g. non-price competition and demand elasticity which is connected with it. A firm offering a niche product faces a different demand elasticity from its competitors. This assumption holds, however, only if the firm specific demand elasticity is constant over time.

36 Sigma_u can be explained as variation between firms, while sigma_e can be explained as a part of variation in

wages which could not be predicted by the specification.

37 “The Hausman test compares two estimators where one is consistent under both H

0 and Ha while the other is

consistent under H0 only.“ (Cameron and Trivedi, 2009, p412) When testing whether to use fixed or random

effects, the random effects estimator is consistent under H0 only. Hausman robust test was used because

homoskedasticity could not be assumed. t-statistics of variables differ by a factor of three between a model not using cluster robust standard errors and the one using them. However, a more formal heteroskedasticity test (Breusch-Pagan test) showed that there is no such problem.

(24)

20

Since the sample period is considerably long in order for demand elasticity to change, this assumption cannot be taken for granted. Nevertheless, the test whether time fixed effects should be used or not showed that time-fixed effect do not need to be included.

Table 4 also reveals that including variables suggested by the literature makes all other variables in the model either less significant or insignificant and reduces the adjusted-R2 considerably. There are most likely two reasons for this. The first one is the reduction in the sample size which gives rise to the violation of the central limit theorem. The second one is as likely as the first one, and probably the result of using wrong variables as proxies for globalisation and technological change. These two processes were recognized by the literature as significant in explaining the wage share of income. But, the lack of additional variables and the inappropriateness of existing variables (R&D cost) for the analysis, constrained the analysis to the use of international sales and capital expenditure as proxies for globalisation and technological change, respectively.

Table 4 Initial regressions - dependent variable: logarithm of unit wage

(1) (2) (3) Overheads 0.05*** (3.39) 0.05*** (3.47) 0.02 (0.90) Sales 0.29*** (5.33) 0.25*** (4.24) 0.16** (2.53) Material costs 0.10*** (3.37) 0.14*** (2.64) 0.10* (2.01) Capital-labour ratio -0.09*** (-3.05) -0.11** (-2.48) Depreciation -0.02 (-1.04) -0.02 (-1.08) Retained earnings -0.00 (-0.25) 0.01 (0.52) International sales 0.11 (1.56) Capital expenditure 0.01 (0.55) Constant 2.17*** (6.81) 2.14*** (4.77) 1.47*** (2.88) Observations 2797 1293 379 Number of firms 152 117 44 R2 0.331 0.394 0.280 Adjusted-R2 0.331 0.391 0.264 sigma_u 1.0406 1.0431 1.0722 sigma_e 0.4014 0.3375 0.2497

All variables represent the natural logarithm of unit variables. Firm fixed effects included in every specification. t-statistic in parentheses. Cluster-robust standard errors used. sigma_u stands for the standard deviation of the individual effect λi, while

sigma_e stands for the standard deviation of the idiosyncratic error εit.

(1) Specification with “Kalecki” variables

(2) Specification with “Kalecki” and “Solow” variables

(3) Specification with “Kalecki”, “Solow” and variables suggested by the literature

(25)

21

The positive characteristic of using logarithmic specification is that coefficients are interpreted as coefficients of elasticity. The second column therefore implies that a 1% increase in firm's overhead costs is expected to increase firm's unit wage by 0.05%.

As mentioned above, Table 4 does not control for the non-stationarity of the logarithm of unit wage. Neither it controls for the different volatility between variables. Thus, the aim of Table 5 is to find an effect of overhead costs on wage share of income that is as exogenous as possible by controlling for all estimation problems discussed above.

In the first four columns of Table 5 each specification is correcting for one estimation problem discussed above which leads to specification in column (4). The specification in column (4) is the preferred one because it employs Driscoll-Kraay standard errors and thereby solves the problem of the unit root in the logarithm of unit wage, the problem of cross-sectional panel dependence and the problem of heteroskedasticity. In the same time it applies moving averages to Overheads, Sales, Capital-labour ratio, Depreciation, Retained earnings and Capital expenditure and thus controls for the different volatility between variables. However, even though this specification controls for all these estimation problems, the Overheads' coefficient does not carry the sign that is suggested by the theory. The coefficient indicates that a 1% increase in firm's overhead costs, increases firm's unit wage by roughly 0.1%.

An interesting feature is noticed in column (2). When controlling for unit root by using the first difference of the logarithm of unit wage as dependent variable, the R2 drops by more than 80% (adjusted-R2 drops even more). This leads to the conclusion that RHS variables explain a very small proportion of the variation in the differenced logarithm of unit wage. Thus, there is even more support for using Driscoll-Kraay standard errors as they automatically control for autocorrelation.

The rest of the table serves as a sensitivity test for coefficients obtained in column (4). An immediate observation is that Overheads' coefficient is not robust at all as both, its magnitude and its sign, change from one specification to the other.

To acknowledge the advice given by Dionne, Gagné and Vanasse (1998) that in the short panel setting it is better to use random over fixed firm specific effects, column (5) uses the same specification as column (2) except that it uses random firm specific effects instead of fixed firm specific effects. Most variables switch the sign in this specification, but it is the only specification that provides a significant Overheads' coefficient with the expected sign.

(26)

22

Table 5 Sensitivity analysis - dependent variable: log unit wage - columns (1), (3), (4), (7), (8) & (10), Δln(wage) - columns (2), (5), (6) & (9)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Overheads 0.024 (0.746) 0.045** (2.393) 0.024 (1.313) 0.097* (2.156) -0.030* (-1.807) 0.047** (2.366) -0.077 (-1.649) 0.097* (2.156) -0.067 (-1.177) 0.009 (0.353) Sales 0.131* (1.937) 0.007 (0.117) 0.131* (1.958) 0.100 (1.228) -0.043 (-1.314) 0.005 (0.081) 0.101 (1.408) 0.100 (1.228) -0.433*** (-3.025) 0.256*** (3.058) Material costs 0.093** (2.071) 0.010 (0.784) 0.093** (2.703) 0.064 (1.059) 0.034* (1.943) 0.012 (0.943) 0.603*** (5.079) 0.064 (1.059) 0.032 (0.933) 0.188*** (6.998) Capital-labour ratio -0.038 (-0.600) 0.024 (0.498) -0.038 (-0.601) -0.372 (-1.630) 0.011 (0.763) 0.019 (0.373) -0.038 (-0.240) -0.372 (-1.630) -0.381* (-1.909) 0.043 (0.595) Depreciation -0.004 (-0.172) 0.008 (0.295) -0.004 (-0.140) -0.034 (-0.317) 0.094** (2.169) 0.007 (0.233) 0.119 (0.987) -0.034 (-0.317) -0.137 (-1.368) 0.124** (2.623) Retained earnings 0.018 (0.613) 0.004 (0.222) 0.018 (0.743) -0.195** (-2.340) -0.005 (-0.265) 0.006 (0.316) -0.111 (-1.261) -0.195** (-2.340) -0.055 (-0.844) -0.102* (-1.781) International sales 0.111 (1.426) 0.034 (1.002) 0.111** (2.358) 0.032 (0.382) 0.028 (1.567) 0.042 (1.043) 0.048 (0.984) 0.032 (0.382) -0.159 (-1.154) Capital expenditure 0.026* (1.820) -0.022 (-1.237) 0.026** (2.219) -0.015 (-0.302) 0.006 (0.224) -0.031 (-1.236) -0.039 (-1.154) -0.015 (-0.302) -0.049 (-0.973) gdp -0.060 (-1.211) 0.006 (0.095) -0.060 (-1.565) -0.053 (-1.073) 0.021 (0.432) -0.021 (-0.339) -0.022 (-0.462) -0.053 (-1.073) -0.046 (-0.823) -0.008 (-0.203) q=2 # Overheads -0.007 (-1.115) q=3 # Overheads -0.005 (-0.808) International sales # moving_cl 0.039* (1.976) Constant 0.842 (1.481) 0.301 (0.649) 0.842** (2.171) 3.492* (1.966) -0.251 (-0.880) 0.180 (0.406) -0.801 (-0.586) -15.122*** (-3.723) -1.618 (-1.006) 1.463** (2.340) Observations 287 286 287 170 170 286 162 170 170 499 Number of firms 43 43 43 32 32 43 29 32 32 81

Avg #obs per firm 6.7 6.7 6.7 5.3 5.3 6.7 5.6 5.3 5.3 6.2

R2 0.246 0.035 0.246 0.171 0.100 0.041 0.435 0.171 0.284 0.384

Adjusted R2 0.222 0.003 -0.178 0.054

t statistics in parentheses. All variables expressed as natural logarithms. Driscoll-Kraay standard errors used in all specifications except in specification (1), (2), (5), (6) & (9)

where cluster-robust standard errors were used. Firm-fixed effects specified in all columns except in column (5) where random-effects are specified. * p < 0.10, ** p < 0.05,

*** p < 0.01. Moving averages of variables that vary on a quarterly basis applied in column (4), (5), (7)-(10). (1) Controlling for GDP growth rate (2) Controlling for the unit

root (3) Driscoll-Kraay standard errors used (4) Overheads, Sales, Capital-labour ratio, Depreciation, Retained earnings and Capital expenditure expressed as moving averages (5) Random-effects regression (6) Interacting Overheads with the quarter indicator (7) Variables trimmed at 1st and 99th percentile (8) Controlling for the amount of total assets firms had when they went public (9) Interacting International sales and Capital-labour ratio (10) Excluding International sales and Capital expenditure

(27)

23

Nevertheless, this specification is less preferred to others as it ignores the fact that firms might posess different characteristics that do not change over time which may drive the obtained result.

Other two specifications that provide a significant Overheads' coefficient as well are the one in which the Overheads' coefficient is interacted with the quarter dummies (6)38 and the one in which all variables are inversely weighted by the value of their assets in the period they went public (8). In the latter, coefficients and their standard errors are equal to coefficients and standard errors in specification (4). The only thing different is a huge and significant constant which suggests that specification fails to include variables that would explain the logarithm of unit wage significantly.

That some variables might be helpful in explaining the logarithm of unit wage is suggested by the R2 in column (7) and (10). In these columns R2 is considerably higher than R2 in other columns which is for two reasons. First, coefficients in other specification are probably driven by outliers (specification (7) includes variables that are trimmed at 1st and 99th percentile). Second, the inclusion of variables suggested by the literature reduces the predictive capacity of the model. The discussion whether these variables should be included or not is deferred to the following section.

In order to test the Stolper-Samuelson theorem, specification (8) includes a variable interacting exactly these debated variables. But it suggests that when a company engages in the world trade, its wage share of income will rise more if it has a higher capital-labour ratio, which contradicts the theorem.

8 DISCUSSION

Specifications in Table 5 did not provide results that were expected by the theory. They also proved that there were some serious limitations in the estimation of results. Therefore, this part will discuss potential reasons why results are different from what was expected, as well as alternative methods that might be employed in order to get results that are more in line with the theory.

The immediate observation from Table 5 is that the effect of overhead costs on wages is in different direction than suggested by the theory. There might be three explanations for such a result. First, the applied theory might not predict real-life data well enough. Second, there are

38 Specification (6) misses one quarter from the analysis (the first quarter is missing to avoid the problem of

perfect multicolinearity), which is due to the inability to perform the analysis on observations which lack data in at least one variable. The lack of Q4 interaction variable means that there is at least one variable that is missing in the Q4 observations.

(28)

24

variables that could bias the Overheads' coefficient in the wrong direction, so a different econometric method should be applied. And third, variables used in the analysis should be structured differently.

8.1 Theory and real-life data

Since the analysis did not estimate the effect of overhead costs on wages in the direction suggested by the theory, there might be other theories that explain the effect better. E.g. it could be that the general positive connection between finance and inequality found by Demirguc-Kund and Levine (2009) is spilled over to the relationship between financialisation and functional income distribution. This might suggest that firms facing larger dividend and interest payouts, were more financially viable to raise more equity or debt which are the reason for larger dividend and interest payouts. Greater financial viability and amount of funds raised may trickle down to wages thus making the connection between overhead costs and wages positive.

Though tempting to accept this explanation as satisfactory, recent developments in financial system, inequality and wage share of income suggest that this explanation does not provide a story that is observed in practice.

8.2 The appropriateness of the econometric method used

Secondly, the econometric method applied might not be appropriate as it might ignore some variables that can bias the coefficient in the wrong direction. The analysis above uses a one-step procedure in estimating coefficients. Thus, there is always a threat that some variables, which are not included in the specification, are simultaneously influencing both, dependent and independent variables. When endogenous variables are recognised, an instrumental variable method might be a better method to apply. However, acknowledging the possibility of endogeneity in overhead costs, this study argues that the instrumental variable method is not appropriate when data is coming from firm’s financial statements. The basis of this argument is inferred from the following example. According to the financial literature, a firm can increase its profitability by increasing its debt level (financial leverage) (Brigham and Houston, 2004). An increase in debt increases firm’s interest payments but the effect of an increase in interest payments on wage share of income is ambiguous. Firstly, as already explained above, Kaleckian theory predicts that an increase in interest payments would have a negative effect on wage share of income. Secondly, an increase in firm’s profits might induce a firm to pay their workers more. In the same time, an increase in debt level might

Referenties

GERELATEERDE DOCUMENTEN

Despite this significant relationship found, no conclusions are drawn based on Table xiii (Appendix V), because Table 8 fails to identify the existence of a relationship

(2011) examine the impact of capital structure on profitability based on 272 American manufacturing firms for a sample period of 2005 to 2007. The findings of this study show

Of course, the level of measured inequality between individuals is higher compared to inequality between households, but the magnitude of the overall reduction in income

As discussed in the theory, it could be expected that women´s job satisfaction increases more than men´s job satisfaction after promotion (Hypothesis 2A) because women

Finally, the third research question we will answer in order to propose an answer to the main research question: “Does this causal relationship between administrative

In the words of the World Bank, “it is to ensure deep, broad and fast debt relief and thereby contribute toward growth, poverty reduction, and debt sustainability in the poorest,

Thus, the companies are aware of their exposure to each of these currencies and they have insulated their positions against foreign exchange risk through the use of

We find that the stringency of measures and fear of the virus had a substantial positive effect on sales and that these relationships were moderated by country and product