Marginal contribution factor
.
The marginal contribution factor tells the % change in bond level over a specific year conditional on lag(log(B/TL)T). Thus the higher (B/TL)T, the smaller the percentage change in level from one year to the next. This conclusion is important to bear in mind as it provides explanation of the results in chapter 5.3. In line with the time contribution factors, the contribution factor for CTT increases over time with a decreasing trend (see marginal yearly change in Table 5.4) and reaches its lowest increase in 2008.
To sum up, the time dummies states that the level never falls during the crisis, but its trend of continuous growth slows down. The cross time-term dummy adds that the same relationship exists conditional on the bond-ratio. An interesting point is that the higher the bond-ratio, the less bond levels differ from its 2003 level. The decrease in the level trend, i.e. the in the marginal contribution factor, reached its lowest point in 2008.
57
5.1.2 Equity
First the regression output is presented. Then it is briefly interpreted. Every time the expression equity-ratio is used in chapter 5.1.2 it is equivalent to the variable lag(log(E/TL). 5.1.2.1 Regression output for equity level
The two sub- regressions used to extract the estimators explain the variation in the equity level with R2 = (Appendix 4.2). This is extremely high and emphasises a high degree of multi correlation.
Table 5.4 provides the outcome, i.e. the estimators for the explanatory variables, of Regression 1 for the logged values of equity level. Thus, equity level is the dependent variable, which is defined as the aggregated value of NFCs’ equity liabilities as quoted at the balance sheet in a specific country. All values are in 2003 $ prices. The equity level is expected to be explained by: 1. The equity-ratio primo the period of interest, i.e. lag(log(ET,C/TLT,C).
In addition the level is expected to vary from year to year, so that the equity level also is explained by exogenous non-stationary macro-economic variables captured by the time dummies (e.g. D2004).
An example could be changes in the Central bank’s monetary policy. It is suspected that the equity level change due to other country specific determinants than just preferred equity-ratio. An example could be GDP or creditor protection. These are exogenous variables captured by the country dummies (e.g. DDenmark). The cross time-term (CTT) captures the part of the yearly changes in equity level that is caused by a non-stationary relationship between the equity-ratio and the equity level. For the time dummies and cross time-terms, year 2003 is used as the reference variable. This means that all estimators represent the change in level compared to 2003. Norway is the country reference variable, and the country dummy estimators reflect how each country’s level differs from the level in Norway. No star= insignificant coefficient, * = Significance at a 10 % level. **= significance at a 5 % level and *** = significance at a 1 % level.
Table 5.4 – Regression I output for equity level
Explanatory variable Estimator Standard deviation
Intercept 5.955*** (.0655)
lag(log(ET/TL)T,C 1.663*** (.1759) Time dummies
D2003 - -
D2004 .009 (.0225)
D2005 .077*** (.0234)
D2006 .138*** (.0259)
D2007 .175*** (.0288)
D2008 .137*** (.0275)
D2009 .207*** (.0239)
Country dummies
DNorway - -
DDenmark -.091*** (.0227)
DSpain -.392*** (.0176)
DEngland .349*** (.0216)
DFrance .355*** (.0184)
DGermany .525*** (.02)
DUS 1.289*** (.0365)
DFinland -.701*** (.0307)
Cross time-terms
CTT2003 - -
CTT2004 .032 (.0603)
CTT2005 -.214** (.0677)
CTT2006 -.466*** (.079)
CTT2007 -.633*** (.0874)
CTT2008 -.468*** (.0643)
CTT2009 -.660*** (.0593)
Source: Appendix 4.2
58
5.1.2.2 Interpretation of the time and cross time-term estimators Table 5.5 – The time estimators for equity level
Time Time estimators Marginal yearly change
2005 .077*** 0.077
2006 .138*** 0.061
2007 .175*** 0.037
2008 .137*** -0.038
2009 .207*** 0.070
Marginal yearly change = DT - DT-1
Source: Table 5.4
All time dummies except for 2004 are significant and carry positive estimators. Marginal yearly change is the change in the time estimators from one year to another. Interestingly the picture is the same as for bonds: From 2005 -2007 the level increased at a decreasing rate. In 2008 the equity level fell back to the level of 2006 (bond level never decreased). Through 2009 equity level again increased and that to a higher level than before the crisis. Equities are riskier than bonds. This could be the explanation as to why the reaction is more drastic than the bond market during 2007. Through 2007 the level increased with 10.037 = 1.088 = 8.8 % and 2008 time the level fell with 1-10 -0.038 = 8.38 %.
As expected, it appears that equity- and bond time factors are highly correlated, and that equity possesses a higher degree of volatility. The reasons for this will be addressed later.
The cross time-term, repeated in Table 5.6 show that the level, regardless to equity-ratio, is increasing every year unless in 2008. For the same reason as for bonds, the level increased more for countries with a low (E/TL)T. One can also say, that the more negative the CTT estimator, the larger a gap will we see between the equity level for market-based countries (high (E/TL)T) and bank-based countries (low (E/TL)T). The reason is that the CTT estimator and lag(log(E/TL) is multiplied when calculating the contribution factor.
Table 5.6 – The cross time-term estimators for equity level
Time Cross time-term Marginal yearly effect
2005 -.214** -0.214
2006 -.466*** -0.252
2007 -.633*** -0.167
2008 -.468*** 0.165
2009 -.660*** -0.192
Marginal yearly change = CTTT - CTTT-1
Source: Table 5.4
59 Until 2008, equity level increased as a function of the equity- ratio and so did the difference in level between market- and bank-based economies. In 2008 the level decreased which is also why the marginal effect is positive in 2008. The decrease was more severe for bank-based economies, i.e. countries with a low equity-ratio.
As for bonds, the CTT contribution factor for equity is extremely interesting because countries with different equity-ratios all things equal will find their equity levels differing more in 2007 than in 2008. The reason is that the factor is calculated by multiplying the CTT estimator and the E/TL ratio.
In practice it means that the countries with low equity- ratios experienced a percentagewise higher increase in level from the glowing economy in 2005-2007. In the same manner they have suffered a larger absolute value decrease in 2008. The marginal effect is used heavily in chapter 5.3.
To sum up, the time and cross time-term dummies are suitable supplements of each other and both show significant results. The cross time-term estimators indicate how much the country levels differ from the one in 2003 depending on the period’s primo equity-ratio and the period in question. The level increases every year, except for 2008, as a function of the equity-ratio.
The level increases less for countries with a high equity-ratio, i.e. market-based economies.
These results are useful for comparison with the corresponding estimators for bond and bank-loans as they together tell which system has been most exposed to the financial crisis.
60
5.1.3 Bank loans
First the regression output is presented. Then it is briefly interpreted. Every time the expression bank loan-ratio is used in chapter 5.1.3 it is equivalent to the variable lag(BL/TL). 5.1.3.1 Regression output for bank loan level
For bank loans no logarithmic variables are included as the observations were rather normally distributed in their natural scale. The result is a regression that is much easier to analyse and that seems more intuitive. The explanation degree (R2) is a bit lower for loans. R2 takes values between 90.59 % and 94.26 %, depending on the sub-regression (Appendix 4.3).
Table 5.4 provides the outcome, i.e. the estimators for the explanatory variables, of Regression 1 for bank loan level. Thus, bank loan level is the dependent variable, which is defined as the aggregated value of NFCs’
bank loan liabilities as quoted at the balance sheet in a specific country. All values are in 2003 $ prices. The bank loan level is expected to be explained by: 1. The bank loan -ratio primo the period of interest, i.e.
lag(BL/TL)t,C . In addition the level is expected to vary from year to year, so that the bank loan level also is explained by exogenous non-stationary macro-economic variables captured by the time dummies (e.g.
D2004). An example could be changes in the Central bank’s monetary policy. It is expected that the bank loan level change due to other country specific determinants than just preferred bank loan -ratio. An example could be GDP or creditor protection. These are exogenous variables captured by the country dummies (e.g.
DDenmark). The cross time-term (CTT) captures the part of the yearly changes in bank loan level that is caused by a non-stationary relationship between the bank loan -ratio and the bank loan level. For the time dummies and cross time-terms, year 2003 is used as the reference variable. This means that all estimators represent the change in level compared to 2003. Norway is the country reference variable, and the country dummy estimators reflect how each country’s level differs from the level in Norway. No star= insignificant coefficient, * = Significance at a 10 % level. **= significance at a 5 % level and *** = significance at a 1 % level.
Table 5.7 – Regression I output for bank loan level
Explanatory variable Estimator Standard deviation
Intercept -150,478 (140,921)
lag(BL/TL)t,C 721,037** (282,670)
D2003 - -
D2004 14,239 (34,062)
D2005 91,596*** (34,953)
D2006 176,471*** (36,531)
D2007 249,129*** (37,701)
D2008 298,143*** (33,947)
D2009 271,789*** (39,558)
DNorway - -
DDenmark -33,425 (38,867)
DSpain 563,532*** (32,084)
DEngland 339,561*** (51,045)
DFrance 292,461*** (52,209)
DGermany 768,419*** (33,828)
DUS 1,133,574*** (117,700)
DFinland -145,004*** (70,929)
CTT2003 - -
CTT2004 45,267 (96,325)
CTT2005 259,575** (104,468)
CTT2006 518,814*** (114,779)
CTT2007 695,826*** (122,226)
CTT2008 670,868*** (95,888)
CTT2009 474,232*** (99,950)
Source: Appendix 4.3
61
5.1.3.2 Interpretation of the time and cross time-term estimators
Table 5.8 – The time estimators for bank loan level
Time Time estimators Marginal Change
2005 91,596 91,596
2006 176,471 84,875
2007 249,129 72,658
2008 298,143 49,014
2009 271,789 -26,354
Marginal yearly change = DT - DT-1
Source: Table 5.7
Interestingly, level changes in BL seem delayed compared to both equity and bond level: The marginal increase for each year slows down through 2007 and 2008, but unlike equity and bond levels, the marginal change (and thus level) is not negative until 2009. It only falls back about one and a half per year, but as we do not have data for 2010 it could decrease further. In this manner, we see a difference between market and bank-based financing, which will be analysed in the discussion chapter.
Cross time-term estimator Χ lag(BL/TL)T, i.e. CTT contribution, states how much the level differs from that in 2003 depending on bank loan ratio. Note that it is not a contribution factor, i.e. it does not express a percentage change in level, but an absolute change. Conditional on lag(BL/TL) T the level is higher in every year following 2003, except 2004.
Table 5.9 – The cross time-term estimators for bank-loan level
Time Cross time-term estimators Marginal yearly effect
2005 259,575** 259,575
2006 518,814*** 259,239
2007 695,826*** 177,012
2008 670,868*** -24,958
2009 474,232*** -19,645
Marginal yearly change = CTTT - CTTT-1
Source: Table 5.7
The higher the lag(BL/TL)T, the higher the CTT contribution. Therefore, any change in the CTT estimator affects the CTT contribution more for countries with a higher BL/TL, i.e.
based-based countries. This is the exact opposite conclusion than what we saw for bond and equity levels. As for CTT Bonds and CTT Equitythe lower the estimator, the more countries with different bank loan - ratios approach each other.
62 This relationship is linear on the original scale, not percentagewise and therefore the finding stands in contrast to those of bonds and equity: The fact that countries with high bank loan ratio have a higher absolute increase in level in a given year, does not mean that they have a higher percentage change in level. However, the percentage change is crucial to estimate, as it enables us to compare bank loan level with equity and bond level. This comparison is done in chapter 5.3.
To sum up, the most interesting finding is that the decrease in bank loan level seems to hit all countries more softly than equity and bonds, and not reach its full effect until 2009. In the meantime bank-based countries felt the change more severely and therefore bank-based countries were more exposed to macroeconomic changes. Hereby I do not claim that the percentage impact is larger in bank-based countries, it could be the opposite for reasons explained later. As the crisis hit in 2009, bank- and market-based levels of loans converged;
the absolute level fell more for bank- based than for market-based countries.
5.2 Research question B
Research question 2 digs a bit deeper and focus on the mix of net-transaction, i.e. the periodical change in stock caused by repayment of old debt/equity and new issuing of debt/equity. Hence net-transaction is the new net-financing for each period cleaned from accounting figure and re-valuation. It indicates the change in finance used for investments.
The focus is not of the absolute change in net-transactions, but instead changes in the mix of net-transactions in form of equity, short- and long-term bonds and bank loans.
The regression tests this question by holding the 6 possible de-components of capital structure up against both country and time dummies. Dummies are used as explanatory variables and the de-components for capital structure as dependent variables.
5.2.1 Results
In appendix 6.1 to 6.6 all the 6 regressions (B-short/B-long, B-short/E, B-short/BL, B-long/B, B-long/BL and BL/E) are presented. In contrast to my expectations, none of them are significant at any level (insignificant p-values, i.e. p-values > 0.1). Therefore none of them are included in the actual paper. To understand why all regressions are statistical insignificant a graph on net-transactions in short-term bond level is shown. Corresponding graphs for the other variables are not presented in the paper, as they tell almost same story (see Appendix 7).
63
5.2.2 Why net-transactions provide insignificant coefficients
The graph below shows net-transaction of all countries except Norway. Norway stands out of any context and adds no information. Its presence would require a scale going down to -10.000. Thus it is excluded from the graph.
Graph 5.4 – Net transactions in short-term bonds
Net-transaction in short-term borrowing equals addition short-term borrowing minus repayments in a given quarter.
All underlying values are quoted in $2003, Q1 prices and the same period is used as Index 100. Source: Domestic statistic databases summed in Appendix 5.
It is crucial to note the tendency of reinforcement characterising all countries, i.e. after a period of increase in net-transactions a period of decrease occurs. It appears as if the observations for each country fluctuate randomly around a mean with no trend. This is exactly why the OLS model finds no significant estimators – there is simply very little system in net-transactions over time. The time dummies typically include 4 quarters of observations. If the switch from increases to decreases in transactions were accompanied with a longer trend, the time dummies would average the four quarter’s opposite signs out and focus on their general trend compared to other years. That this is not the case confirms what look like a random walk.
-600 -400 -200 0 200 400 600 800
Index 100 = 2003, Q1
Net-transaction in short-term bonds
France Germany US Spain Denmark Finland England
64 Surely, companies do not issue securities or require additional loans in the bank based on pure randomness. On the other hand, decisions on new financing do not have to follow a systematic trend over time. For example, a given company might have a predetermined idea about how projects are refinanced and such an idea is not necessarily something that is changed every year or month. In addition, and probably more importantly, investments are seldom something that is planned to be done in predetermined intervals. Sometimes good investments turn up and sometimes they do not. Good investments occur more for conservative companies in bad times where prices are low. For more risky companies investments can be low in bad time, because they are out of capital, whereas the same companies heavily invested in good times. Therefore, many reasons for holding a certain net-transaction level for all types of finance exist and these reasons are not necessarily systematical over time or for countries in between.
5.3 Research question C
When comparing the estimators on the cross time-terms with estimators on equity and bank loans it is possible to see which market that changes most from year to year. Holding the estimators against each other while keeping country specific characteristics constant, highlights which market that was most affected by the crisis and thus which system that lost the most while keeping country specific characteristics constant. Question C asks whether absolute level changes and structural changes in net-transactions differed depending on the type of financial system. The conclusions from research question A and B works as the base arguments for answering research question C. As the results of Regression II were insignificant they are only used to support arguments related to the findings in Regression I.
Research question C will therefore only use findings from Regression I.
Sub-chapter 5.3.1 elaborates on the properties of the marginal contribution factor, as it is the key to estimate what type of financial system that is mostly affected by the recent crisis and how. Sub-chapter 5.3.2 presents the marginal change in level over the year for each country and each type of external finance conditional of the capital ratios (e.g. the bond-ratio). Sub-chapter 5.3.3 weights the findings in sub-chapter 5.3.2 and estimates the total change in external finance for all countries during the crisis.
65
5.3.1 Mathematical understanding of financial system-differences
The cross time-term, must be seen as a variable allowing the impact of the capital-ratio on the capital- level to change in a non-linear way over time. However, within a given year, the cross time-term still provides a linear relationship and this is true for all types of external finance.
For equity and bonds, such relationship is percentagewise linear. For bank loans the relationship is linear in absolute values. The total change in level over time conditional on the capital -ratio is calculated by multiplying the capital - ratio with the CTT estimator at a given point in time. This is referred to as the cross time-term’s contribution to level. It is calculated for all types of finance. For bonds and equity contribution regards the logged levels. For bank loans contribution regards the level measured in the natural scale. For bonds and equity a contribution factor was calculated, because it enabled us to focus on contribution of the cross time-term to the level measured in the natural scale.
The Marginal yearly change was in chapter 5.1 calculated as the difference between the CTT estimator T and the CTT estimatorT-1. The marginal contribution factor for bonds = % ∆ in the contribution factor, . And the marginal contribution factor for equity = % ∆ in the contribution factor, .