Effects of Stock Market Liquidity on Growth: Empirics and Theory

Department of Economics

Uppsala University

Master’s Thesis

Author: Martin Dalsenius Supervisor: Dr. Annika Alexius Spring semester 2007

Effects of Stock Market Liquidity on Growth:

Empirics and Theory

Abstract

Historically, it has been difficult to obtain solid data on stock market liquidity for large parts of the world. In recent years, however, the availability of data has improved, but does still have troubles with concentrations to recent years and to relatively wealthy countries. Thus, samples based on balance between industrialized and developing countries may easily come to include very few of the poorest countries. Also, recent theory suggests differences in the impact of stock market liquidity on growth between very poor and other countries. While several papers that have used these data have found significance for liquidity on growth globally, they have also been criticized for potential selection bias and for limited time depth.

In this thesis we construct a sample based on attempting to get the representation of the poorest countries at least reasonably good. We then test the significance of liquidity on growth globally using this sample; we test if there are significant differences between countries of different wealth levels and if the time span is still long enough for the results of our tests to be relevant.

We find that adjusting for the under representation of the poorest countries comes at the expense of getting a

very limited time span. The sample became strongly weighted to the mid and late 90s and the impacts of large

unsubstantiated fluctuations in global stock prices (the Asian crisis) are evident in the regression results. We

conclude that as of today the data simply are not good enough, or extend far enough back in time, for estimating

the impact of liquidity on growth globally, or differences between groups of countries, over a relevant time span.

Abstract ... 1

1.Introduction: ... 3

2. Theory ... 6

2.1 Differences between Debt and Equity:... 8

2.2 Is equity financing equally important regardless of the general wealth levels of an economy? ... 9

3. Earlier studies:... 11

4. Method and Measurements: ... 13

4.1 Measurements: ... 15

5.Results: ... 17

6.Conclusions: ... 22

Resources: ... 24

Appendix 1: Mathematically deriving the formula for Blackburn et al´s critical capital level (below which only debt financing is feasible)... 25

Appendix 2: Country list tables for the selection bias measurement. ... 28

1.Introduction:

Whether the level of development of financial markets affects growth, and in that case how, is something that has been debated on a theoretical ground for quite some time. But historically it has been all but impossible to empirically verify any theoretical models on a global scale.

The data, if not the financial markets as such, simply did not exist for large parts of the world.

However, this is something which is about to change.

In recent years, the World Bank has systematically measured and recorded global data over time, not only of various GDP related measurements and variables traditionally associated with growth regressions, but also on a number of financial indicators and added those into easily accessible databases (WDI, GDF etc. see references). This access to data has spurred a new awoken interest in trying to estimate the affects of finance on growth empirically on a global scale (rather then just for a few wealthy industrialized countries of which data was easier to come by).

However, there have also been some controversies as to the usefulness of the data on many financial indicators as it is today. This is because the series of data on many such indicators still do not really extend all that far back in time, and because data for many of the poorest countries often are systematically scarcer and more concentrated to the most recent years then data for more developed countries are. Choosing your sample countries after the quality of data thus might come with the risk of getting the whole sample strongly biased towards richer countries, while not doing so might mean getting a sample with too many observations

concentrated in too narrow a time period.

One particularly interesting area where such controversies have arisen is that of measuring the

impact of stock market liquidity on growth. It is interesting because liquidity measurements

directly measure the degree of activity of stock markets, and so give a very close proxy of

transaction costs and how efficient the markets for equity are in general (as opposed to for

instance Stock Market Capitalization which only measures the size of the stock markets and

not really if any trade goes on there, or volatility which measures the stability of the stock

markets). Hence if we find that stock market liquidity has a significant influence on growth,

we also find that stock market efficiency impacts growth. But this is provided that we use an unbiased sample with a reasonably large time span.

As noted above, several earlier studies (for instance Levine & Zervos 1998, Levine et al 2000, Beck & Levine 2002 and Aghion et al 2005) have found significant positive impact of stock market liquidity on growth, whereas others (Trew 2006 and Driffill 2002) are skeptical to some of the implications of these earlier studies and raise concerns about the limited time depth and the concentration of data to richer countries and more recent years in the data series used as well as the potential for problems with selection bias. Furthermore relevant theory suggests that the efficiency of stock market liquidity in promoting growth (when measured as the usefulness of equity financing) should be significantly diminished in economies with less access to capital then in economies with more access to capital (see our later theory

discussion). This makes the possibility of selection bias, with respect to systematically under representing the poorest countries (because they have the scarcest data and the shortest time depth for their data) particularly troublesome.

In light of this, the purpose of this thesis will be to see if stock market liquidity is still significantly correlated with growth, for the world as a whole, when making sure that the sample used does not under represent the poorest economies too grossly, and examine

whether it is possible to create such a sample without having the time depth of the sample and the average number of observations per country drop too much. We will also examine

whether there are significant differences between the impact of stock market liquidity on growth in poor, middling rich and rich countries (as defined by their average Real GDP/Capita for the 1960-2005 period).

Obviously the issue of selection bias is relevant for this investigation. However, we will not attempt to formally estimate selection bias, either for the samples used in previous studies or for our own. This is because the core to the problem of the selection bias in this study is, very specifically, the degree to which the poorest economies are included in the samples used

,

1

Because the distribution of available data are such that choosing “a balanced sample of developing and industrial countries” (which is something of a standard procedure for many growth regressions, but may be unfortunate in this case) would be very likely to systematically exclude the poorest countries if one chooses

“developing countries” after the quality of the available data. But perhaps even more importantly because

excluding those countries would mean systematically excluding countries where theory (see our later theory

section) suggests that the impact of liquid stock markets on growth may differ significantly from the rest of the

sample.

rather then the degree to which a sample as a whole can be considered representative for the total population. We find that the most direct way for measuring the degree of representation of the poorest countries is simply to measure how big a fraction of the included countries in a sample that comes from the poorest fraction of a certain size (in our case we use 1/3) of the total population of countries. It should be noted for the rest of this thesis, when we discuss selection bias, how to estimate and compensate for it and so on, this refers to the specific problem of under representation of the poorest countries.

Also, we will use a relatively simple model consisting of two different measurements of stock market liquidity, Value Traded liquidity

and Turnover ratio (the reason why we include two measurements is described later in the Method and Measurements chapter), Bank Credit ( in order to be able to distinguish the impact of liquid stock markets from that of an efficient loans market) and a measurement of Initial GDP level (which is important both because of the significance attributed to differences in wealth levels between countries in this thesis and because it is commonly considered to have a large impact on growth) as lagged independent variables and Real GDP/Capita growth as the dependent variable. The reason we do not want to make the model too complex is that the time depth of the available data is narrow in the earlier studies we will start by examining and is likely to get worse if we must adjust the samples to include more poorer countries (with fewer observations more concentrated to recent years). The risk of over interpreting findings from a very limited time period is thus very real, and the more complex our model the greater that risk becomes.

Since the concepts behind why the efficiency of stock markets may affect growth are essential for this thesis we will begin, by a theory summary of that subject. Then there will be a literary summary of relevant research and surveys on topics that are significant to this thesis and also a general discussion on the methods and measures used in the later empirical part of the thesis.

After that we will conduct our own empirical investigation to solve the issues described in the purpose

. Finally we will make some conclusions to sum up this thesis.

2

More complete definitions of the measurements are included in the latere Method and Measurements chapter.

2. Theory

Why should one expect that liquid stock markets might be important for promoting growth in the first place?

. Garcia&Liu, (Garcia&Liu, 1999:32), identify 3 main channels through which a well developed financial sector may enhance growth:

1) Through promoting savings and providing savers with higher yield alternatives.

This is possible because financial intermediaries and markets can accumulate expertise utilize economies of scale and make those available to the public.

2) Through reducing information and transaction costs in the financial markets. This helps funneling capital between borrowers and lenders more easily, but may also play a part in reducing the costs of asymmetric information between managers and stockholders within the same corporation

3) Through a more efficient allocation of recourses.

Point 1 is pretty straight forward: Liquid stock markets provide savers with efficiently priced higher yield/ risk alternatives to debt saving and should overall provide savers with higher utility and returns to savings

.

Point 2 is really more ambiguous, because while liquid stock markets may reduce information costs and can be used as an instrument for keeping agency costs under control

, they do the latter at a cost. The minimum agency costs (0) occur where there is only one combined long term manager/owner. But the reduced observability in more diversified and short termed ownership structures, that are likely to occur in liquid stock markets, will cause the potential

4

whether or not this also leads to higher savings rates does, however, depend on whether the income or substitution effect, with respect to present and future consumption, dominates (see Beck & Levine, 2002:1)

5

The idea here is that market capitalization of stocks, for instance through dividend policy, will provide all

potential investors with more reliable information on the performance of the company (verified by the actions of

large lenders and underwriters of stock) then might be obtained directly from managers. (see Easterbrook, 1984)

for agency costs to increase (see Barney et al. 1985:25-27) Hence it is far from certain that more liquid stock markets will lead to reduced agency costs.

But it is really point 3 that is most essential for our analysis of why theoretically finance, in our case the liquidity of stock markets, should affect growth. Why is there a need for liquid stock markets with low transaction costs, where equity can be traded easily and how does this lead to a more efficient allocation of recourses?

One thing about stock markets is that, usually, direct issuing of stock where a company sells its shares to a first hand holder of stock stand for a very small percentage of the total trades that go on there. Hence it might seem to a casual eye that the often speculative trading forth and back of stocks have little to do with any efficient allocation of recourses that might promote growth. But equity is essentially an infinite stream of uncertain incomes and most investors have finite time horizons for their investments. The expected existence of an

efficient market where the investor can trade the equity off for an efficient price at the time of his choosing is therefore extremely essential for how the investor will value the equity at the time of purchase. Only then will each holder of stock from the initial issuer and on be able to expect getting the optimal utility of the equity and therefore be willing to pay accordingly. If inefficiencies and high transaction costs can be expected further down the line, they will affect the price that the initial issuer of stocks can get for the equity.

So if a company wishes to use equity to finance its projects, then an efficient allocation of resources can occur only if the stock markets are efficient (liquid).

But instead of issuing equity a company might choose to issue debt (or, very frequently, a

combination of the two). Both equity and debt financing are solutions to the problem of

acquiring present funding in exchange for uncertain future income (even debt is uncertain,

because there are, for instance, bankruptcy costs). But there are some important distinctions

between the two that suggest that the absence of efficient markets for equity will matter for

the efficiency of acquiring present funding and so impact growth:

2.1 Differences between Debt and Equity:

When a company issues equity, it offers rights to shares of its total future expected value (including dividends, the value of the company’s assets and the value of power over the decision making process in the company that comes with the shares) in exchange for

immediate funding. When instead it issues debt, it offers an interest and a payback schedule in exchange for immediate funding.

Because the value of equity over time is always perfectly correlated with the performance of the company over time

the risk, but also the part of the potential benefits of the companies projects, that come with the purchase of equity is carried entirely by the purchaser. When you purchase debt the only risk you take is that of bankruptcy, on the other hand you have very little to gain if the company performs unexpectedly well (except that this decreases the possibility of bankruptcy). The company’s risk is however considerable: it must pay the fixed costs of interest to the debt holders regardless of how its projects turn out and the debt costs will decrease the company’s margins and increase the risk of bankruptcy. On the other hand debt may be tax deductible.

As a consequence of this the preferences for both companies and investors when it comes to choosing one sort of financing or the other (or a combination) may vary a great deal

depending on the situation. They cannot be presumed to be always perfectly exchangeable for one another, despite that they solve similar problems

Hence, liquid stock markets provide means for using equity financing efficiently, and equity financing is an important complement for debt financing when companies need funding for new projects. This would indicate that liquid stock markets should provide a more efficient allocation of recourses and therefore promote growth.

But can we expect the degree of growth promotion to be the same everywhere or does that depend on, for instance, the development level of an economy?

6

Except for the potential influence of agency costs, that is.

7

This is presuming that markets are imperfect and so are influenced by, for instance, bankruptcy costs and tax

incentives. In a famous study by Modigliani and Miller (Modigliani and Miller, 1958) they conclude that all

differences between using debt or equity financing can be derived from such market imperfections.

2.2 Is equity financing equally important regardless of the general wealth levels of an economy?

Allthough, without more thorough empirical investigations this is hard to determine with certainty, it has been frequently suggested that liquid stock markets, as opposed to efficient banks, may be less efficient in countries with a low development level then in countries with a higher one.

Driffill (Driffill, 2002, p.13-17), suggest that part of the reasons for greater inefficiencies in developing countries or countries going through large economic transitions (such as many central and east European countries of the 1990s, when market economy and stock exchanges were reintroduced there) may be derived from the legal system, both the letter of the law in some cases (this may take the form of insufficient protection for minority shareholders, inequality between the rights of different investors (for instances on the basis of whether they are foreign or national) etc), but more frequently from how the law is upheld and the trust and expectations that investors consequentially are able to put in the stock markets. If, for

instance, the control mechanisms of the financial markets are poor and lots of illiquid

securities that may be very difficult to distinguish from solid liquid ones beforehand, are listed on it the efficiency of stock markets will be impaired.

Driffill (Driffill, 2002, p. 4) also mentions the possibility that in imperfect financial markets,

where agents have limited (or imperfect) access to finance, wealth and income distributions

may affect economic growth. The idea here is that there may be some sort of threshold

amount of capital needed for many investments and that those who have that amount of

capital readily available will be able to achieve a higher rate of return then those below it. If

we take this to include both debt and equity financing, it becomes rather obvious that having

readily available capital does matter. Both types of financing carry costs, as we have seen

earlier in this thesis, if nothing else the time and effort needed to convince investors of the

profitability of the projects the firm wishes to undertake added to the risk premium that

investors will charge for being unable to perfectly distinguish between the profitability of

different projects.

But in fact, it has been argued that having a large part of the population very poor will impact not only general investment levels but also the optimal capital structure of firms in the

economy. Blackburn et al (Blackburn et al, 2003) find that, up to a certain level of

development (actually available capital in the economy), only debt financing is meaningful whereas after that level a combination of equity and debt financing becomes preferable, they even derive an approximate formula for a critical level of capital that marks the border between the two

.

For the purpose of this thesis, the possibility of a profound difference between the usefulness of equity financing in rich and poor countries, is of course of huge importance.

Consequentially studying not only the world as a whole, or even its regions separately, but also dividing countries into groups after their general GDP level will be important when trying to find statistically significant influences on growth from efficient stock markets. We will return to this in our later discussion on method and measurements.

8

The argument goes that in sufficiently capital poor economies it is virtually always worthwhile for investors to

invest the labour time necessary to control the choice of projects themselves, leaving only the issue of effort

related agency costs unresolved. Whereas in richer countries this is not necessarily the case. A downsized

version of the mathematical calculations they perform to derive their formula has been included as appendix 1.

3. Earlier studies:

A number of recent studies (Levine et al, 2000; Levine & Zervos, 1998; Beck & Levine, 2002; Aghion et al, 2005) find significance for the impact of various financial indicators including stock market liquidity on growth. Those studies use panel data based on the Worldbank databases (WDI & GDF). Aghion et al (2005) find support for strong economic convergence among highly liquid economies, in a survey where they use the same data set as Levine et al (2000).

Other studies (Driffill, 2002; Trew, 2006) question the plausibility of some of the implications of the above studies.

Trew also addresses the problems of selection bias and too limited time frames (Trew, 2006:

3-7) . One big reason for those problems is the lack of older data for a great many developing countries. If we use series of data that only cover recent years we risk overinterpetation of temporary events, such as the Asian Crisis of the late 90s and the results may thus be less relevant then those of studies that cover longer time series. But if we wish to study longer time series and wish to include countries that have consistent data for the whole time period then we risk ending up with a very biased selection of almost exclusively developed

countries. In Trew´s review of previous studies (Trew, 2006:3-7) he points out that in the

Levine & Zarvos study (Levine & Zarvos, 1998), the number of missing observations during

the early part of the period they investigate is systemetically larger then in the later part and

that the missing data systematically comes from poorer countries. Trew summarises the

variation of missing data over time in the following graph:

(source: Trew, 2006:6)

4. Method and Measurements:

Overall, given the available data, it will be difficult to avoid that the number of missing data will be systematically concentrated to the first decades and to developing countries, unless one wishes to use a very strong selection of countries where data are better and/or cover a very limited period of time. In the first case the sample is extremely likely to be strongly selection biased (very unrepresentative) and strongly tilted towards developed countries, in the second case temporary phenomena and deviations may be given grossly too much influence, so that once again the study becomes unrepresentative.

When we conduct our own empirical investigation we will, in order to come up with an unbiased sample, first create a tool for estimating selection bias: we simply divide the

countries available from the World Bank database World Development Indicators Online into three categories after average Real GDP/Capita for the 1960-2005 period and check for too large an under representation of the poorest group

. We begin by applying this tool to samples used in some previous studies that found a strong significance for finance on growth. If we find no selection bias in those samples

, we will accept the best sample and increase the time span to include the years of the Asian crisis and see if we still find significant results for the impact of liquidity on growth.

If we do find selection bias we will proceed by setting up a full sample of available economies from the World Bank database World Development Indicators Online, and include an

economy into a data pool as long as it has at least two observations

. While this strategy will not completely irredicate the problems of selection bias, since the economies that lack

observations (and therefore have to be excluded) are very likelly to systematically be poor

9

It should be observed that these are averages of the available data for the period. Some countries did not exist in 1960, a few have scattered or incomplete data for the year 2000 US $ GDP/Capita. But those are very few, the data series are far and wide better then for the liquidity measurements, also for poor countries.

10

As we want to test for the possibility that equity financing is systematically less efficient in the poorest countries (in accordance with Blackburn et al) we will be rather strict and call any sample that doesn’t have at least a 25% ratio of countries that are among poorest 33% selection biased. This is an arbitrary measure, using a relatively arbitrary tool, but it should give a general idea.

11

While using two, or very few, observations to represent an economy may in each individual case allow for

quite a big margine of error, the idea is that with a sufficient number of such economies added into the data pools

developing countries, it is the best possible strategy to minimize them. This strategy does however also have a big disadvantage: it will not be possible to use moving averages to cancell out business cycle effects when we measure financial market data, the way Beck &

Levine (2002) do. That would simply make it neccesary to remove too many economies from the study .

Based on the best sample, we will then run several regressions to test if we can find a significant impact of liquidity on growth for the world as whole and also for the different wealth categories. For the world as a whole we will test both the full model (with both liquidity measurements, bank credit and initial GDP level as independent variables) and each of the financial market measurements (liquidity measurements and Bank Credit) separately together with initial GDP level as well as initial GDP level in isolation to test impact on growth. This is to determine how the different financial market measurements separetelly affect R

and to test how well the full model stands compared to its parts.

For the different wealth categories we will only run the full model as this suffices to see if there are any significant differences between the different wealth categories. We will also investigate whether country specific constants have an important effect by running all models both with and without those included in the regressions (whereas when we run the regressions without them we assume one constant for all instead).

After that the results will be summarised and we will write our conclusion.

4.1 Measurements:

As mentioned above we will divide the countries of the world into 3 categories after their wealth as measured by average Real GDP/Capita for the 1960-2005 period.

We use two different measurements, Turnover ratio (the value of the trades of domestic shares on domestic exchanges divided by the value of domestic shares) and Value traded liquidity (the value of the trades of domestic shares on domestic exchanges divided by GDP) to measure the liquidity of the stock markets. The reason why we need to use both Turnover ratio and Value traded liquidity as measures of liquidity is that they measure quite different aspects of it.

Turnover measures the value of trades of domestic shares on domestic stock exchanges divided by the value of listed domestic shares, thus providing a good measure of transaction costs in the domestic stock exchanges. Value traded on the other hand measures the value of the trades compared to the GDP of the economy as a whole, including non listed firms and might thus give better measurement of the transaction costs of equity trading (and issuing) in the economy as a whole. Value traded Liquidity, however, does have a potential pitfall namely that: “If markets anticipate large corporate profits, stock prices will rise today. This price rice would increase the value of stock transactions and therefore raise Value Traded.

Promlematically, the liquidity indicator would rise without a rise in the number of

transactions or a fall in transaction costs.” (Levine & Zervos 1998:8-9). However, Turnover is not effected by this price effect since it has stockprices both in the numenator and

denomenator. Therefore, we can use the relationship between Turnover and GDP economic growth to check for inconsistencies between it and the relationship between Value Traded Liquidity and economic growth. That way we can check for the posibility that the price effect might be dominating the relationship between Value Traded Liquidity and economic growth.

Hence it would make sense to include both variables, even thouth Value Traded Liquidity will

be the primary measurement.

Our growth measurement, the dependent variable, is Real GDP/Capita growth, throughout this thesis. We do not use multiple-year averages on this variable. Doing so would mean removing more years of data for our liquidity measurements and that would increase the problems of bias further. While this means the dependent variable is not compensated for business cycle variations, it also means that if a single business cycle has too large an

influence (which would be an indication of that the time span measured is too small) this will show in the regression.

Besides the stock market liquidity variables we also use the following variables:

Bank Credit (as defined by domestic credit to private sector) to measure the efficiency of the debt markets.

Moving averages of year 2000 US$ GDP/Capita levels 10-15 years before as a measure of initial GDP level.

Real GDP/Capita growth as the dependent variable.

All independent variables (except Initial GDP level as mentioned above) will be lagged 5 years to compensate for causuality problems.

When we use the country specific constants we assume a model of:

Y

= β1

*X1

+ β2

*X2

+ β3

*X3

+β4

*X4

+ε

:

where Y

= Real GDP/Capita growth, X1

– X4

are the independent variables: initial GDP level, Bank Credit, Value Traded Liquidity and Turnover ratio, respectively. β1

– β4

are the slopes for each independent variable’s correlation with the dependent variable. ε

is the standard error. We also have country specific constants for every country used.

When instead we assume the same constant for all it becomes:

Y

=C

+β1

*X1

+β2

*X2

+ β3

*X3

+β4

*X4

+ε

With no country specific constants but a constant C

for all instead.

5.Results:

First we divided the original 186 countries and economies of WDI online into three groups after average real GDP/capita for the period 1960-2005. The groups of countries included in each category are included in a table in appendix 2.

Using the ratio of countries belonging to the poorest category we found that, although 18 of the 40 countries in the Beck & Levine study (Beck & Levine, 2002)

are “developing countries” only 4 of them belong to the poorest 1/3 of countries and economies available in WDI. In another study that uses a similar “balanced mix” of industrialized and developing countries (Mauro, 2002) that number is 2 out of 31. Hence the initial suspicion that merely dividing between industrialized and developing countries might easily lead to selection bias with respect to wealth level conditions can be considered to be confirmed.

We therefore proceeded by making a sample including all of the above countries and economies that had at least one observation for both Value Traded Liquidity and Turnover (and for Real GDP/Capita growth) and ended up with a somewhat reduced sample also included in appendix 2.

Quite clearly the countries or economies removed become systematically more frequent as we move down in wealth group. To be more precise we have that of the 62 poorest countries 39 have to be excluded because they lack any data whatsoever. For middling rich countries the figure is 26 out of 63 and for the richest countries 14 out of 61. This means that 23 out of 107 countries and economies included in the regressions come from the bottom 1/3 in our sample.

This is a significant improvement in reducing the selection bias

, and we still have some significant results for our liquidity measurements as can be seen in the tables below. But they

12

Although the Levine & Zervos (1998) and Levine et al (2000) studies are more frequently quoted, both in the reference literature and this thesis, the Beck & Levine (2002) study is in many ways a more updated version of the previous two, that seeks to address some of the statistical problems encountered in those. We will therefore use the sample from that updated study, rather then the samples from the older studies, as a reference point for the empirical part of this thesis.

13

But it should be remembered that the countries and economies excluded are still the ones with the worst data

availability and so, by the same argument as earlier, are likely to have been systematically different from the

sample norm if it would have been possible to measure them. Furthermore, it would have been possible to create

a sample which gave each wealth category exact proportionate influence, by removing some of the wealthier

countries from the sample, but it was decided not to do so simply because this would weaken the influence of the

are not all that many and none of the full model regression lack at least one strongly insignificant variable.

None:

Model 1 Model 2 Model 3

Model 4 Model 5 Lagged

Real GDP/Cap

1.24*10^- 5

(11.6)

4.03*

10^-5 (3.23)

9.19*

10^-5 (7.85)

7.48*

10^-5 (4.18)

4.98

*10^-6 (0.19)

Bank Credit

1.7*

10^-2 (10.6)

1.27

*10^-2 (2.94)

Value Traded Liquidity

1.40*

10^-2 (4.92)

-1.22

*10^-2 (-1.70)

Turnover ^1.88*

10^-2 (7.85)

2.67

*10^-2 (6.08)

R2,adj -4.21% -2.38% -10.7% -9.04% -3.43%

Fixed:

Model 1 Model 2 Model 3

Model 4 Model 5 Lagged

Real GDP/Cap

-1.84*

10^-4 (-5.05)

-1.92*

10^-4 (-5.23)

1.89*

10^-4 (-1.59)

3.10*

10^-4 (-1.50)

-1.52*

10^-4 (-0.514)

Bank Credit

6.56*

10^-3 (2.93)

-5.38

*10^-2 (-4.16)

Value Traded Liquidity

5.73*

10^-3 (1.25)

2.94

*10^-3 (0.259)

Turnover ^1.18*

10^-3 (0.265)

-1.07

*10^-3 (-0.134)

R2,adj 11.2% 11.6% 19.7% 12.3% 12.8%

concentrated to an even narrower time span (as countries with very few observations very systematically have

them concentrated in the most recent years), then the sample used. The chosen sample is in other words the result

of a compromise.

None:

Rika länder

Mellan Fattiga Lagged

Real GDP/Cap

8.83*

10^-5 (2.22)

9.12*

10^-4 (3.95)

2.01*

10^-4 (0.103)

Bank Credit ^4.86*

10^-3 (-0.646)

-9.77*

10^-3 (-1.21)

4.78*

10^-2 (1.88)

Value Traded Liquidity

1.84*

10^-2 (-2.14)

1.31*

10^-2 (1.01)

-4.95*

10^-2 (-0.815)

Turnover ^3.06*

10^-2 (4.91)

7.99*

10^-3 (0.833)

1.63*

10^-2 (1.60)

R2, adj 0.925% 0.801% 10.2%

Fixed:

Rika

länder Mellan Fattiga

Lagged Real GDP/Cap

4.22*

10^-5 (0.131)

-4.40*

10^-3 (-3.37)

-1.21*

10^-2 (-1.55)

Bank Credit ^-3.23*

10^-2 (-1.25)

-4.48*

10^-2 (-2.76)

-6.29*

10^-2 (-1.17)

Value Traded Liquidity

5.81*

10^-3 (-0.337)

1.21*

10^-2 (0.721)

3.61*

10^-3 (4.79*10^-2)

Turnover ^1.41*

10^-2 (1.02)

1.51*

10^-2 (1.11)

-2.94*

10^-2 (-2.29)

R2, adj ^1.34*

10^-2 %

22.0% 36.9%

That, in the tables above, R

values are generally higher when we use country specific constants in our regression then when we use only one constant for all is perhaps not so surprising, that after all should diminish the residuals. Nor That Lagged Real GDP/Capita and Bank Credit have a greater significance for impact on growth then do our liquidity

measurements. That is well in line with previous studies.

What is surprising is that in every regression run Turnover Ratio has a greater significance then Value Traded Liquidity. Unless there is some great external chock that causes

unsubstantiated fluctuations on global stock markets (large unsubstantiated fluctuations in stock prizing would, as mentioned in the Method and Measurements chapter above, affect Value Traded Liquidity and impair its usefulness for measuring liquidity, but this would not affect Turnover) there is no logical explanation for this. After all, under perfect market conditions, the performance of Value Traded Liquidity and Turnover over time can only differ significantly if large shifts in how big a part of the total economic activities of a country that occurs in companies and firms that are listed on the stock markets occur. Even in that case it is only if having an extremely small but highly liquid stock market somehow becomes an optimal point for affecting growth that Turnover can be more significant. The latter is of course nonsense, so the only realistic explanation is that of a huge chock to world stock prices in the measured period that would impair the usefulness of the Value Traded Liquidity

measurement.

Another thing with the regression results above is that we cannot really clearly distinguish the expected differences between poor and richer countries when it comes to the impact of liquid stock exchanges on growth. For Turnover Ratio it seems almost the other way around, that it would be more significant for poorer then richer countries and even though Value Trade Liquidity seems to follow expectations better in this sense, it should be noted that the R

values drop dramatically when we move from the poorer to the richer countries in both tables where we make this division above. The possibility of a chock to the global stock markets, and the global economy have a large influence on our measured period would help explain the inconclusiveness of these findings too.

If, when comparing our sample to that of Beck and Levine (2002), we take a look not only at selection bias but also at time depth and average no of observations/country, we get a hint at the answer to what that chock might be.

Value Traded liquidity, using the Beck &

Levine, 2002 sample

Value traded liquidity, our sample

average no. of observations/country 12.8 9.40

ratio of countries with observations

before 1990 97.5% 47.6%

ratio of countries with observations

before 1994 100% 62.9%

Turnover, using the Beck & Levine,

2002 sample Turnover, our sample average no. of observations/country 8.75 6.52

ratio of countries with observations

before 1990 40.0% 18.5%

ratio of countries with observations

before 1994 85.0% 49.0%

Beck & Levine, 2002 sample our sample ratio of countries belonging to the

poorest 1/3 10.0% 21.5%

It is very clear from the table above that the improvements in selection bias when switching the Beck & Levine sample for ours has come at the cost of narrowing the time span

significantly, both for Value Traded Liquidity and Turnover, and also reducing the no of average observations. As 5-year lagged variables are used for all independent variables (to compensate for causality issues) in the regression, all observations on liquidity after 2000 fall outside the scope of this study. That means that our regressions are extremely weighted towards the mid and late 1990s.

Incidentally that period is, somewhat simplified, characterized by the periods before and during the Asian crisis. The period before the Asian crisis, being one of a strong global growth, not least in economies with an efficient, liquid financial market, and the Asian crisis period, which incidentally often hit those same economies with liquid financial markets the hardest.

6.Conclusions:

Considering the data on Stock Market Liquidity it is obvious that a number of things can be remarked upon when making a study based upon such a scattered set of data so concentrated to a very limited time period. For one thing, this study has (as opposed to most other studies on similar subjects) to a large degree focused on attempting to reduce the potential selection bias by including as much data as possible, also from countries and economies that are normally excluded because of their scarcity of data.

The rationale for this has been that if scarcity of data on stock market liquidity cannot be assumed to be a random variable, but is rather likely to be strongly correlated with both low GDP level, low economic growth rates and general political instability

, then by including only countries for which there are reasonably good sets of data we get a sample that is strongly tilted towards wealthier countries with more ordered civic conditions. This would clearly be undesirable even if there was no specific reason to suspect that poor countries would differ fundamentally from wealthier ones when it comes to the impact of liquid stock markets on growth (extrapolating the available data too far beyond the interval which is actually measured is not a good scientific procedure). In this case the theory provided by Blackburn et al (Blackburn et al, 2003), does suggest that there might be systematic differences between the poorest countries and the richer ones.

While other studies (for instance Beck & Levine, 2002 or Mauro, 2002) have dealt with the selection bias issue by trying to get a reasonably balanced sample of developing and

developed countries that have reasonably good data available, we have found that measure unsatisfactory in this case. “Developing country” is a rather arbitrary term after all, and if we choose developing countries after the quality of available data, then the sample chosen is likely to systematically exclude the poorest, slowest growing economies. Instead we

determined what countries were to be considered poor bases on averages of Real GDP/Capita

14

It would be quite reasonable to assume that very poor countries with very low economic growth, possibly

combined with political instability, normally have a less developed (or at least less well functioning) financial

sector and almost certainly are more likely to have less reliable, or consistent, data on the performance of the

financial sector over time. The data on which the empirical part of this thesis is based strongly support that the

missing data is concentrated to the poorest countries.

for a longer time period. In this way we got a sample that, while no doubts less selection biased also has a greater number of observations concentrated to the most recent time period and a lower number of average observations/ country overall. We also found that a great deal of the available observations, particularly for poorer countries, were concentrated to the period before and during the Asian crisis.

By contrast, the Beck and Levine (2002) study, ends in 1998. Levine et al (2000), Levine &

Zervos (1998) and Aghion et al, all end in 1995 and so would not be able to fully incorporate the effects of the Asian crisis. Furthermore, those studies have been much more selective in choosing countries with better data (and therefore systematically use very few of the poorest countries which have their data most concentrated to the latest part of the period).

But although, very likely, the results of our regressions could have become very different if we had ended up with data covering a different time period (or if there had been no Asian crisis), that may not be so important anyways. With too many single observations

concentrated on too short a time frame, it is far too likely that that short time frame would be unrepresentative for the longer time series that we cannot see (because the data are lacking).

So in fact we have only managed to trade selection bias problems for problems with too

limited a time depth and while it may be possible to trade the two forth and back it will not be

possible to create an unbiased, relevant sample with the available data. If the world bank

continues to record financial and growth data in the same pattern as in recent years, then

maybe in another 10 or 20 years the data series will have enough time depth and cross

sectional breadth to make a relevant investigation on the impact of liquid stock markets on

growth, but not yet.

Resources:

Aghion, P., Howitt, P., and Mayer-Foulkes, D. (2005). “The Effect of Financial Development on Convergence: Theory and Evidence”. The Quarterly Journal of Economics, 120(1). 173-222.

Barnea, A.,Haugen, R.A. & Senbet, L. W., ”The nature of Agency Problems, Agency Problems and Financial Contracting, Prentice-Hall Inc., 1985.

Beck, Thorsten & Levine, Ross., “Stock Markets, Banks and Growth: Panel

Evidence”. Working paper 9082, National Bureau of Economic Research, July 2002 Blackburn, Keith; Bose, Niloy & Capasso, Salvatore. “Financial Development, Financing Choice & Economic Growth”. Working Paper no. 96, Centre for Studies in Economics and Finance, Dipartemento di Scienze Economiche-Università Degli Studi di Salerno, 2003

Driffill, John. “Growth and Finance”. School of Economics, Mathematics and Statistics, Birkbeck College, University of London, 2002

Easterbrook, F.H. “Two Agency-Cost Explanations of Dividends”. The American Economic Review, september 1984

Garcia, Valeriano F. & Liu, Lin. “Macroeconomic Determinants of Stock Market Development”. Journal of Applied Economics, Vol. 2, No. 1 (May 1999): 29-59

Levine, Ross; Loayza, Norman; & Beck, Thorsten. “Financial Intermediation and Growth: Casuality and Causes” Journal of Monetary Economics, No. 46, 2000:31- 77

Levine, Ross & Zervos, Sara. ”Stock Market, Banks, and Economic Growth”.

American Economic Review, June 1998: 537-58

Mauro, Paolo. “Stock returns and output growth in emerging and advanced economies”. Journal of Development Economics 71 (2003) 129-153

Modigliani, F. and M. H. Miller, 1958, The Cost of Capital, Corporation Finance and the Theory of Investment, American Economic Review 48: 261-297.

Trew, Alex William. “Finance and Growth: A Critical Survey”. University of St Andrews, Centre for Dynamic Macroeconomic Analysis Working Paper Series No.

CDMA05/07. Revised April 2006.

World Bank Databases:

WDI: (http://ddpext.worldbank.org/ext/DDPQQ/member.do?method=getMembers&userid=1

&queryId=135/)

GFD:( httphttp://ddpext.worldbank.org/ext/DDPQQ/member.do?method=getMembers&useri

d=1&queryId=136/ )

Appendix 1: Mathematically deriving the formula for Blackburn et al´s critical capital level (below which only debt financing is

feasible)

The following appendix is a downsized version of the calculations that Blackburn et al perform (Blackburn et al 2003:9-17) to which the author has taken the liberty of adding some comments of his own. All formulas are however taken directly from Blackburn et al

(Blackburn et al 2003:9-17) and appear in the same order as in that text.

They use a two period Overlapping Generation Model consisting of households (financiers of risky projects and suppliers of labour when young and consumers of final output when old) and firms (operators of risky projects when young and employers of labour and producers and consumers of final output when old). The firms try to optimize their profit given the available financing and financiers try to optimize their earnings given the available projects to invest in.

Neither the effort level nor the actual projects each firm produce are observable to the

households by default, but it is possible for the households to eliminate the uncertainty about project choice through spending 1-η units of its own labour time in acquiring information and securing the project of its choice.

Blackburn et al assumes the following capital function:

l

= units of loans, h

= units of entrepreneurial time, x

is a project specific value.

From this they derive the formula for the expected value of capital to:

In addition they assume that entrepreneurs may choose to spend a certain amount of their time (1- h

) converting their own labour into output and that running a project. With a probability of q an entrepreneur who runs a project obtains an output of φ(1- h

) whereas an output of φ

is obtained by an entrepreneur who does not a project. It is assumed that φ > φ due to

positive externalities (knowledge and spillovers) and those externalities are assigned the value Ф. Furthermore it is assumed that an old entrepreneur employing n

units of labour and k

units of capital is able to produce:

K

denotes the aggregate stock of capital.

Given that the firm is bound to pay d

≥0 in debt payments and s

, 0≤ s

≤1 in equity payments and that the interest rate is r we have that the utility functions of a firm and a household, respectively, are:

If the household chooses to eliminate the cost of not being able to observe the company’s choice of project at the expense of 1-η units of its own labour time we now face the optimization problem:

which is solved by:

however, if the household chooses not to intervene in the choice of project we have that:

and must further add the restriction that:

this gives quite different solutions to the maximization problem:

Notably, only debt financing will be used if households choose projects for themselves whereas a combination of debt and equity is optimal where firms do this for them.

Entering the different solutions into the household utility functions give us the utility that households have of a debt-only and a debt-plus-equity contract, respectively:

From this we can then derive an approximate formula for an critical level of capital, k

,

below which pure debt-only contracts are preferable and above which mixed debt-equity

contracts are preferable:

Appendix 2: Country list tables for the selection bias measurement.

Selection bias groups:

poorest 62 Middle 63 Richest 61 ETHIOPIA HONDURAS ST. LUCIA BURUNDI PHILIPPINES SLOVAK REP MALAWI SYRIA ESTONIA

NEPAL IRAQ GABON

GUINEA BISSAU CONGO, REP POLAND MOZAMBIQUE UKRAINE CROATIA ERITREA DJIBOUTI LIBYA CHAD MOROCCO MEXICO BURKINA FASO CAPE VERDE SEYCHELLES

MALI EGYPT LEBANON

UGANDA BOLIVIA URUGUAY NIGER SERBIA MALTA CONGO,

KINSHASA BOSNIA KOREA, REP RWANDA NICARAGUA ST, KITTS SIERRA LEONE ALBANIA

CZECH REPUBLIC GHANA SWAZILAND OMAN TANZANIA THAILAND VENEZUELA TOGO GEORGIA

TRINIDAD &

TOBAGO CAMBODIA ECUADOR TAIWAN

LAOS SAMOA PALAU

BANGLADESH VANUATU PORTUGAL INDIA PARAGUAY ARGENTINA BENIN BELARUS ANTIGUA TAJIKISTAN KAZAKHSTAN GREECE

CENTRAL AFRICA TONGA CYPRUS SUDAN TUNISIA SLOVENIA MADAGASCAR GUATEMALA PUERTO RICO GAMBIA

WEST BANK

AND GAZA SPAIN VIETNAM

DOMINICAN

REP SAUDI ARABIA LESOTHO BOTSWANA BAHRAIN

KYRGYZ REP IRAN NEW CALEDONIA SAO TOME COLOMBIA NEW ZEALAND GUINEA BULGARIA SINGAPORE NIGERIA ALGERIA

FRENCH

POLYNESIA

MAURITANIA EQ GUINEA ISRAEL

PAKISTAN FIJI IRELAND

KENYA MACEDONIA MACAO

TIMOR LESTE JORDAN ITALY

COMOROS NAMIBIA BAHAMAS

CHINA ROMANIA HONG KONG

MONGOLIA EL SALVADOR AUSTRALIA

SENEGAL ST. VINCENT BELGIUM ZAMBIA RUSSIA FINLAND BHUTAN MICRONESIA FRANCE INDONESIA PERU NETHERLANDS YEMEN, REP BELIZE AUSTRIA LIBERIA MALDIVES

UNITED KINGDOM SRI LANKA MALAYSIA CANADA MOLDOVA SURINAME GERMANY PAPUA NEW

GUINEA TURKEY ARUBA ZIMBABWE

MARSHALL

ISLANDS SWEDEN UZBEKISTAN BRAZIL ICELAND KIRIBATI MAURITIUS DENMARK HAITI DOMINICA ISLE OF MAN ARMENIA JAMAICA NORWAY

CAMEROON LATVIA UNITED STATES SOLOMON

ISLANDS COSTA RICA JAPAN

ANGOLA CHILE LUXEMBOURG COTE D IVOIRE SOUTH AFRICA KUWAIT TURKMENISTAN PANAMA SWITZERLAND AZERBAIJAN GRENADA UN ARAB EM GUYANA HUNGARY

LITHUANIA

Our sample:

poorest 62 Middle 63 Richest 61

_MALAWI _PHILIPPINES _SLOVAK_REP _NEPAL _UKRAINE _ESTONIA _UGANDA _MOROCCO _POLAND

_GHANA _EGYPT _CROATIA

_TANZANIA _BOLIVIA _MEXICO

_BANGLADESH _SERBIA _LEBANON

_INDIA _SWAZILAND _URUGUAY

_KYRGYZ_REP _THAILAND _MALTA

_NIGERIA _GEORGIA _KOREA__REP_

_PAKISTAN _ECUADOR _CZECH_REPUBLIC _KENYA _PARAGUAY _OMAN

_CHINA _KAZAKHSTAN _VENEZUELA__RB

_MONGOLIA _TUNISIA _T_AND_T

_ZAMBIA _GUATEMALA _TAIWAN _INDONESIA _BOTSWANA _PORTUGAL _SRI_LANKA _IRAN _ARGENTINA

_MOLDOVA _COLOMBIA _GREECE

_PAPUA_NG _BULGARIA _CYPRUS

_ZIMBABWE _FIJI _SLOVENIA

_UZBEKISTAN _JORDAN _SPAIN

_ARMENIA _NAMIBIA _SAUDI_ARABIA

_COTE_D_IVOIRE _ROMANIA _BAHRAIN _GUYANA _EL_SALVADOR _NEW_ZEALAND

_RUSSIA _SINGAPORE

_PERU _ISRAEL

_MALAYSIA _IRELAND

_TURKEY _ITALY

_BRAZIL _HONG_KONG

_MAURITIUS _AUSTRALIA

_JAMAICA _BELGIUM

_LATVIA _FINLAND

_COSTA_RICA _FRANCE

_CHILE _NETHERLANDS

_SOUTH_AFRICA _AUSTRIA

_PANAMA _UNITED_KINGDOM

_HUNGARY _CANADA

_LITHUANIA _GERMANY

_SWEDEN

_ICELAND

_DENMARK

_NORWAY

_UNITED_STATES

_JAPAN

_LUXEMBOURG

_KUWAIT

_SWITZERLAND _UN_ARAB_EM