Page 1
1
Shadow Banking as a Catalyst for Real Growth
Academic year 2012-2013
Assistant: Reina Renard
Promotor: Prof. Dr. Kristien Smedts
MASTER OF FINANCIAL ECONOMICS
Thesis submitted to obtain the degree of Master of Financial Economics
Maarten Gielis
FACULTY OF ECONOMICS AND BUSINESS
s0193708
Page 3
3
TABLE OF CONTENTS
1 INTRODUCTION .............................................................................................. 7
2 LITERATURE .................................................................................................. 9
3 REVIEW OF SHADOW BANKING ................................................................ 11
4 MODEL .......................................................................................................... 14
5 DATA ............................................................................................................. 16
6 RESULTS ...................................................................................................... 18
6.1 REGRESSION ANALYSIS ................................................................. 18
6.2 ROBUSTNESS TESTS ...................................................................... 20
6.3 COMPLEMENTARY REGRESSIONS ................................................ 22
CONCLUSION .................................................................................................. 23
APPENDIX ........................................................................................................ 25
REFERENCES .................................................................................................. 27
Page 4
4
Acknowledgements
I would like to thank my thesis supervisor, Reina Renard, for her time and efforts in order to
help me with my master paper. She contributed a great deal to this paper by giving several
writing tips. Moreover, she guided me through my problems and answered all my questions
in a clear way. I am also very grateful to my promotor, Prof. Dr. Kristien Smedts, who gave
me the opportunity to write my thesis about a very interesting and topical subject.
Page 6
6
Shadow Banking as a Catalyst for Real Growth
Abstract
The shadow banking (SB) system has existed over 50 years and has increased enormously
the past ten years. Moreover, the total market value of SB activities is estimated at 60 trillion
USD worldwide. In spite of these facts, the research on this subject is scarce. This paper
tries to complement the existing literature by investigating the influence of SB activity on real
economic growth. The former is approximated by leverage growth of SB flows, the latter is
measured by GDP growth. Our main hypothesis is that SB activity positively affects GDP as
the financial intermediary system provides credit to institutional investors. This in turn,
increases investments and hence potential GDP. The impact of SB is estimated by means of
a multivariate regression analysis with SB leverage as main variable. The model is then
complemented with some control variables that have been proven to have a real influence on
GDP. Our main hypothesis is thus not confirmed in the data. This might encourage the
authorities to monitor the SB system more closely and impose certain restrictions on financial
innovation. Indeed, in this case, SB activity has practically no influence on economic growth
but is still a potential danger for the economy due to increased risk and complexity. Hence,
the disadvantages outweigh the benefits of the SB activities.
Page 7
7
1 INTRODUCTION
Shadow banking (SB) activities and the entire credit intermediary system supporting these
activities are described in detail by Pozsar et al. (2010). This is an important first step in the
process of understanding and defining SB. As there is no clear definition for SB, a standard
definition helps one to understand the functions that SB activities perform. Moreover, it can
disclose the possible dangers that they contain. The term shadow banking was first used in
2007 by Paul McCulley, a managing director at PIMCO. This expression contains a
pejorative meaning for SB activities and might give the wrong impression to those
concerned. It is certain that the shadow banking system can be of great value for the market
as a liquidity provider and risk sharing mechanism. In contrast to possible gains, SB activities
played a large part in the financial crisis that started in 2007. In short, the high leverage and
maturity mismatch led to an inability of providing liquidity when needed in the market. The
costs, a decrease in asset value accounted for 13 trillion dollars for the USA (Better Markets,
September 2012).
This research paper focuses on the real effects of SB activity and hence the importance of
SB throughout the years. The main hypothesis is that increasing SB activity has a positive
impact on the real economy. Indeed, the SB activities are an alternative source, compared to
traditional banking activities, of credit for investors. There is a clear relationship between the
strength of the financial system and economic growth. One observes the more developed
countries to have a large financial system. As more credit is available, the economy is
stimulated through their investments. Jobs are created which increases consumption and in
turn booms economic activity. This market mechanism ensures real growth. This relationship
is empirically investigated in the literature. King and Levine (1993) find that the size of the
financial system, measured by the relative size to GDP and the amount of credit given, is
strongly correlated with economic growth. In order to investigate the relationship between SB
activities and real growth, a multivariate regression is used to estimate this effect. Therefore,
this study is based on the findings of Banerjee and Marcellino (2005) for the multivariate
regression and more specifically for the choice of the explanatory variables. The analysis
includes both macro-economic and financial variables that are found to predict economic
growth. This paper tries to complement previous studies by assessing the additional
information the growth of the SB system contains in estimating the economic activity.
As far as we are aware, no research has been done on the empirical relationship between
shadow banking activity and real growth. Therefore, this paper contributes to the scarce
empirical literature on SB as well as to the literature that predicts economic growth. We
Page 8
8
follow the research of Adrian, Moench and Shin (2010) who find a significant relationship
between asset prices and SB leverage. A logical extension of their research is to investigate
the predictive power of SB activity on the economy. Hence, a better understanding of the role
of SB in the economy is obtained through our research. This is important as SB activities
comprise 40 percent of the total financial system and continue to grow further. On the one
hand, as the contribution of SB proves to be significant, regulators must be aware of its
benefits when implementing certain restrictions on SB activity and thus leave more room for
innovation and financial reform. On the other hand, when the impact of SB on real economic
activity turns out to be small, the authorities might find the downsides of SB to exceed the
possible benefits. In this case, strict monitoring and regulation is the best solution to prevent
a financial crisis in case of failure of the SB activities. Moreover, this paper is of practical
importance as the public demands more insight in the role of financial intermediaries and
their interconnectedness with the real economy, since the financial crisis that started in 2007.
The multivariate regression with GDP as dependent variable, SB activity as main explanatory
variable, and eight other macroeconomic and financial control variables, we obtain a model
with a strong explanatory power. Most control variables have significant influence on GDP
and their signs are as expected. In contrast, our main hypothesis that SB positively impacts
economic growth is not confirmed. In fact, the coefficient for the SB proxy is significantly
negative. However, when testing the robustness of these results, we find that this sign and
this significance do not hold. Hence, the data cannot confirm a significant relationship
between SB activity and real growth.
The definition by the Financial Stability Board (FSB) that is broadly used for shadow banking
distinguishes between SB entities and activities. SB is in fact the total of non-bank financial
intermediation. Some entities that are included in the SB system are money market funds
(MMF), special purpose vehicles (SPV) and investment funds that are leveraged. SB
activities are, among others, securitization, securities lending and repurchase transactions.
The total value of the SB system has been estimated by the FSB at 60 trillion dollars in 2007,
of which half is the US market. This represents some 35 percent of the total financial system.
SB is similar to traditional bank activity, in the sense that it also intermediates through credit
creation. Indeed, liquidity is created by transforming loans into assets on the basis of maturity
and liquidity. Loans, e.g. mortgages are issued and therefore financing is needed and can be
achieved by selling securities. Hence, SB is characterized by very high financial leverage.
However, SB activities are not regulated and are hence not provided with support from
central banks or the public sector i.e. the government. Moreover, the function of SB is also
quite different from regular banking because whereas the latter has access to funding
through deposits, the former needs to sell securities to institutional investors in exchange for
Page 9
9
collateral. These securities are created by the different players, conform to their liquidity and
risk needs. Several types of loans are pooled, tranched and enhanced to get to different
classes of securities. Basically, the purpose of SB is to make profits by borrowing short term,
liquid, creditworthy securities and issuing long term, risky loans.
The remainder of this paper is organized as follows. The next section discusses the literature
related to this research paper i.e. SB literature and literature on forecasting growth. Section 3
reviews the background of shadow banking and its different aspects more in detail. The
focus is on the advantages, disadvantages and current discussion points for regulation. The
fourth section explains the model which is used to predict economic growth and discusses
the rationales for each variable included. Section 5 then describes which data are used for
each variable and also, the methodology of this research. Section 6 gives the results of the
regression analysis and provides some basic intuition to these outcomes. Some robustness
tests are performed to help interpret the results. Thereafter, complementary regressions are
executed that further assess the power of the regression analysis. Finally, in the last section,
we give some concluding remarks.
2 LITERATURE
Although the shadow banking system exists some 50 years, there has not been much
research on this subject until 2000. From that time on, SB activities have enormously
increased in size which attracted attention not only from the financial world but from
researchers as well. The first articles on SB were published in 2008. Some papers have
been published on the functioning of SB, as well as on the role of the SB system in the
financial crisis. Pozsar et al. (2010) give a definition of the SB system and assess its scope.
They define SB as a whole of financial non-bank intermediaries. The authors value the SB
system by two measures. The gross measure of SB is a summation of all securitization
activity liabilities, whereas the net measure attempts to exclude double counting. In order to
compute the latter, they use the Flow of Funds provided by the Federal Reserve. The
research of Pozsar et al. (2010) attempts to describe the whole credit intermediation process
as well. In short, the SB system is a chain of different activities and entities. The first step of
this process is loan origination, which is performed by finance companies. These loans are
then put together by conduits, so-called loan warehousing. Broker-dealers structure these
loans into term asset-backed securities. Finally, structured investment vehicles and credit
hedge funds conduct the intermediation of ABS. All these steps are funded in the wholesale
funding markets. The authors note that not all steps are used as this is dependent on the
Page 10
10
quality of the originated loan. Adrian, Moench and Shin (2010) empirically investigate the
predictive power of balance sheet and macroeconomic variables on excess returns for
several portfolios. They use the Flow of Funds data to examine the relationship between
asset prices, balance sheet adjustments and real growth. They use a large set of equity
portfolio returns, credit returns and Treasury returns as dependent variable. Their research
shows that two balance sheet variables namely the annual growth of security broker-dealer
leverage and the quarterly growth of SB total financial assets are best in predicting returns.
The evidence is proven robust, as complementary robustness tests give similar results.
Looking at these results, it is thus interesting to examine the role of SB variables on real
growth, as they are able to predict real returns.
Further literature on SB is rather scarce. This limited availability is mostly due to the lack of
data. Most published papers focus on the description of the SB system, its advantages and
disadvantages and possible solutions to regulatory issues. Also, much research focuses on
the failure of the SB system which has contributed a great deal to the financial crisis of 2007.
Pozsar (2008) also explains the SB system. He distinguishes between the risk originators
and risk bearers as the main players and describes the different steps from loan origination
to final securities intermediation i.e. the asset and funding flows. These steps are as
described earlier in this paper. Amery and Van Deyck (2012) write their dissertation on the
SB system in which they describe the evolution and functioning of the SB system and its
advantages. Moreover, they address possible issues by reviewing monitoring efforts and
regulation.
Since the analysis of the SB as a predictor for real growth is based on a multivariate
regression, it is interesting to examine which variables should be included in the model. First,
Marcellino (2007) examines the predictive power of different types of time series models for
forecasting GDP and inflation. He compares the quality of a linear and a non-linear model
with simple autoregressive models. If simple linear models are correctly specified, they are
difficult to beat i.e. they have the best predictive power for the lowest complexity. However,
the author suggests that findings of a linear model should be compared to a non-linear
model, as these often give other results. This gives more robust conclusions of the analysis.
As for which variables must be included in a GDP forecasting model, our paper is based on
the findings of Banerjee and Marcellino (2005). Their research focuses on finding leading
indicators for GDP and inflation for the Euro-area. They perform an analysis using Euro-
indicators as well as US macroeconomic variables. Indeed, they expect the impact of the
USA on the Euro-area to be substantial. As in the previous paper, linear models are
compared with simple autoregressive models. They find that the autoregression has lower
RMSE – a measure of estimating error – than 50 percent of the indicator models in 13 out of
Page 11
11
16 periods. However, the best indicator outperforms the AR every time though this best
indicator is different in most periods. The results for the Euro-variables show that among the
best single indicators are short and long term interest rate, the unemployment rate and
government investment variables. They repeat this exercise for the US-variables and find
that industrial production, unemployment rate and the consumer sentiment are among the
best predictors in a single indicator model. Again, it is important to keep in mind that these
best-performing variables change over time. One can hence conclude that a forecasting
model should be updated regularly. Another paper by Camba-Mendez et al. (1999) uses a
so-called Automatic Leading Indicator (ALI) model that dynamically selects the best
indicators from a set of variables. This avoids the arbitrariness of randomly selecting a
number of indicators. GDP is forecasted for four European countries. Among the most
frequently selected variables by the ALI model are the short term interest rate, the share
price index and industrial production. More importantly, they find that their ALI model always
outperforms autoregressive models in at least half of the cases. This implies a great value for
dynamic models in forecasting real growth.
3 REVIEW OF SHADOW BANKING
In this section, the shadow banking activities are discussed in detail. The first part handles on
the scope of the SB activities. Next, we provide the reader with more insight in the
advantages and disadvantages of SB. Thereafter, some regulatory issues are reviewed. This
discussion is important to assess whether or not SB should be monitored and constrained in
order to prevent a financial crisis while still preserving the possible benefits. Of course, one
should be aware of possible supervisory and regulatory activities that already have been
proposed or implemented.
Pozsar et al. (2010) estimate the total value of the SB system. At the end of 2007, when both
SB leverage as well as regular banking leverage had increased to its highest points in
history, SB gross and net liabilities were around 21 and 17 trillion US dollars respectively.
This is an increase in estimated net liabilities of more than 400 percent as compared to 1990
figures. Traditional banks had liabilities worth 11 trillion US dollars in 2007. Moreover, the
FSB estimated the global SB system around 60 trillion US dollars in their Green Paper on
SB. From this, half is represented by the US market.
The SB system is advantageous for many reasons. First, it provides liquidity to the financial
market and through this, to the real economy. Funding is provided by SB activities even
Page 12
12
when the traditional banks are not able to invest in the real economy through loans. This
implies that an increase in liquidity and capital requirements encourages the growth of the
SB activities. Loans are originated as a third party needs financing. These loans are being
pooled and sliced, and several types of securities are created. Examples of such securities
are asset-backed commercial paper (ABCP), asset-backed securities (ABS) and
collateralized debt obligations (CDO). As a result, the supply of liquidity is of such a level that
it fits the demands of the investors i.e. the buyers of the security. To create a sufficiently
large supply, assets are transformed into securities tailored to the needs of the buyer. Then,
the originators are able to sell these short-term and less risky assets in a liquid market.
Another advantage is that risks are managed and diversified better. On the one hand,
securitized assets can be moved of the balance sheet and on the other hand, revenues are
improved without changing the balance sheet structure. This kind of diversification moves the
risk away from the banking balance as risky, long-term, less transparent assets are
securitized into the liquid balance sheet items which reduces risk.
However, SB has several downsides as well. In case of runs of the system, the
intermediaries are not able to meet the liquidity demands. This causes a negative spiral such
that the financial system and the real market crashes. The reasons for this are the presence
of systemic risk in the market and the complexity of the SB system. Because of the securities
created by SB activities, the financial intermediaries became highly leveraged. In fact, this
leverage was hidden in the sense that securitization enables SB to put assets off the balance
sheet. Hence, investors willing to buy these kinds of securities were not fully aware of the
risk. They expected their risk to be lower than it actually was. More importantly, the whole
securitization process is one of the root causes of the financial crisis. When the US housing
market crashed, the securities resulting from SB activity decreased rapidly in price as these
were linked to mortgages. Indeed, the combination of junk loans and the drop in housing
prices caused the mortgages to lose value. As these mortgages were securitized, the prices
of these securities decreased. Due to the interconnectedness, the whole financial system
collapsed, collateral was sold with losses, which led to even more securities failing. The run
on the SB activities contributed to the scarce liquidity in the financial market which also
affected the traditional banking system. Indeed, all the credit risk that was put off balance
sheet needed to be taken on by the traditional banks. These junk securities created losses
after the assets were marked-to-market.
In order to tackle these issues, or prevent them if possible, several reports have been
composed since the start of the financial crisis. The European Commission (2010) describes
the SB system and states briefly its advantages and disadvantages before proposing certain
measures to monitor and regulate the SB system. Following the FSB, they suggest the
Page 13
13
following three-step approach to understand, monitor and regulate the SB activities. First, the
relevant entities and activities in the SB system should be identified and monitored closely.
Even though SB has long existed and is more present than ever, the data are still
incomplete. The data collection may be complicated by the fact that intermediaries reside in
a foreign country that impedes gathering the necessary data. In a next step, it is important to
accurately determine an appropriate approach to regulate the SB by filling the gaps where
current regulation lacks in avoiding SB systemic risk. Also, it should be proportionate and be
integrated with available frameworks. In a last step, one has to decide to which extent the
regulation should be enhanced. Indeed, the monitoring framework should be flexible and
adaptable to capture financial innovations. In short, the FSB wants to eliminate the systemic
risk while holding on to the benefits of SB i.e. providing credit and diversifying risk.
Besides recommendations by the European Commission, we find also suggestions in the
literature to improve the SB system. A paper by Adrian, Covitz and Liang (2013) suggests
the following supervisory activities. First, one has to keep in mind several characteristics of
systemic risk. These events have a direct effect on the real economy through contagion and
financial interconnectedness. Indeed, when the shadow activities are hit and vulnerable to
runs, traditional banking institutions provide them with funding. In order to prevent systemic
risk, supervision must be on a different level besides the microprudential level. An
assessment process is suggested in order to implement several scenarios that are used to
identify possible shocks on SB. Next, the transmission channels through which these shocks
are amplified must be examined. From this, the influence on real activity should be assessed.
The authors warn that systemic risk might emerge during boom periods as it is built up during
periods with low volatility. Another danger that is mentioned is the increased aggregate risk
through financial innovation. Innovation allows for risk sharing, but in a high risk environment,
this increases aggregate risk. Second, the authors argue why and how the SB system should
be monitored. As SB is highly leveraged, asset price increases are much larger in boom
periods. More importantly, price crashes during recessions are amplified as well. The SB
activity should be assessed by using several measures of risk. The authors propose to
constantly monitor possible price bubbles, high SB leverage and vulnerability due to
mismatches in maturity and credit.
This paragraph is based on a presentation by El Khadraoui (2012) and it brings some
additional discussion points of the benefits and dangers of SB activities upon which the
reader might reflect. As the SB system provided more liquidity in the market, traders found it
opportune to speculate in the period before the financial crisis. This problem of moral hazard
increased the bubble which eventually led to the market crash. It is interesting that a market
should be liquid but might be too liquid on a certain level. This can lead to irrational behavior
Page 14
14
of market participants. Secondly, as mentioned before, the supervisors should implement
general regulation rather than specific. This allows for flexible changes as is necessary for a
rapidly changing environment. Last, the proposal of overregulation of the SB system might
be detrimental in the long run. It is clear that the non-bank financial intermediaries lack the
appropriate supervision. However, this may be extended to the extreme case of the 1980s
when a Belgian bank was not profitable and needed to increase margins in order to survive.
In this view, SB should only be regulated to a certain extent as the financial system needs to
be profitable such that this sector remains stable.
4 MODEL
The purpose of this paper is to investigate to which extent the changes in SB activity act as a
catalyst for growth in real GDP. The contribution of this paper to the literature is the
assessment of the influence of the SB activity on real growth and the additional predictive
power that goes with it. To investigate this relationship, the setup of the model is
straightforward. A multivariate regression is estimated with real GDP growth as dependent
variable. The choice for a multivariate regression with variables other than the SB variable is
clear as previous studies showed the predictive power of certain financial and
macroeconomic variables. The GDP is a measure of real activity as it sums the net monetary
value of all the produced goods and services. Specifically, it is the sum of consumer
spending, business investment, government spending and trade deficit. Whereas nominal
GDP is a pure summation, and hence does not account for price increases, real GDP
removes the impact of inflation and is thus lower than nominal GDP.
A number of explanatory variables for explaining GDP are considered, but the main one is
the growth in SB activity. Others are the unemployment rate, the short-term and long-term
interest rate, inflation, industrial production, etc. These are discussed below. In defining SB,
the literature does not agree on the scope of the SB system. However, on the definition of SB
there is more consensus, although there are also differences. For instance, Pozsar et al.
(2010) devote their paper to researching and defining the SB system. They state that SB
contains the activities and entities that provide liquidity in the non-commercial market and
that the core activity is the credit creation through different kinds of transformation of non-
transparent, risky, long-term assets into short-term, liquid liabilities. Note however that the
use of short-term liabilities to fund long-term assets is a risky activity, since it is prone to
liquidity risk. In our paper, the definition of Adrian, Moench and Shin (2010) on SB is used to
assess total SB activity. Hence, we sum over three types of intermediaries: finance
Page 15
15
companies, funding corporations and asset-backed security issuers. As we try to determine
the impact of growth, we use growth in SB leverage. Leverage is defined as the ratio of total
assets divided by equity with SB equity measured as the difference between total assets
minus total liabilities. Higher leverage is associated with higher amount of credit loans and
this in turn increases market liquidity. We expect that a higher SB leverage has a positive
influence on GDP since the market is more liquid and more credit is provided and thus one
can extend their business by increasing investments. Another way to measure SB activity is
the growth in SB assets and SB liabilities. Indeed, the size of the financial system is
positively related to economic growth (King and Levine, 1993). Therefore, several models are
tested with each of the main explanatory variable assessed in a different model.
As this paper tries to estimate a multivariate regression, some control variables are needed
in the model to predict GDP. More specifically, the following variables are included: the
unemployment rate, the short-term interest rate, the rate of certificates of deposits, inflation,
the long-term interest rate, the term spread, growth in industrial production and changes in
the money supply M2. The choice for these variables is supported by the research of
Banerjee and Marcellino (2005). They investigate which indicators are best in forecasting
GDP for the Euro-area. They find significant results i.e. outperformance of the autoregression
for these variables in certain periods. As a result, this paper incorporates these variables in
the regression as well. First, the unemployment rate is used as a explanatory variable.
According to Okun’s Law (1962), the unemployment rate is very closely related to the level of
GDP and they state that an increase in unemployment goes hand in hand with a decrease in
GDP. Moreover, based on intuition, it is quite straightforward to see that the lower the
unemployment rate, the higher the potential output is, and hence the more value created.
Second, one can expect inflation to have a negative effect on future real activity. With higher
prices, the available money cannot buy the same amount of goods and services as before
the price increase. Related to this, rising oil prices also have a detrimental effect on GDP. As
the economy is still largely driven on fossil energy sources such as crude oil, an increase in
the price of this energy source slows down activity. In contrast, an increase in industrial
production and hence in potential productivity benefits real growth. As a fourth and fifth
variable in this analysis, we look at the yield curve and the interest rate. Dotsey (1998)
investigates the predictability of these factors on GDP growth. He finds that the interest rate
and the term spread have good predictive power. More specifically, one can expect a
negative relationship between the interest rate and GDP growth. As higher interest rates are
detrimental to investment, real growth decreases. Moreover, a downward sloping yield curve
is an indication of the present time being an economic boom, and predicts a recession. We
expect thus a positive predictive relationship between the yield curve and GDP. The previous
Page 16
16
findings are related to an increase in the money supply, our sixth variable. It is measured by
the M2 parameter of money supply. M2 equals all savings deposits and non-institutional
money-market funds plus M1, which includes all physical money. Expansionary monetary
policy i.e. a rise in M2 leads to a decrease in interest rates, which in turn increases the
investment and hence stimulates the GDP growth. Table 1 sums up the expectations for the
sign of the coefficients, as can be derived from the argumentations above. This may be
useful in the later analysis of the regression results.
VARIABLE
CODE IN REGRESSION
EXPECTED
EFFECT ON GDP
SB LEVERAGE CHANGE FLOWLEVCHANGE +
SB SIZE GROWTH SB_ASSETGROWTH +
UNEMPLOYMENT RATE CHANGE UNEMPLOYRATCHANGE -
ST INTEREST RATE THREEMONTH_COD -
LT INTEREST RATE THIRTY_IR -
YIELD CURVE THIRTYTHREESPREAD +
MONEY SUPPLY MTWOCHANGE +
INFLATION INFLATION -
ENERGY PRICE OILPRICECHANGE -
PRODUCTION IPCHANGE +
Table 1 Expected effect on GDP of several explanatory variables
5 DATA
In order to obtain the data for the empirical analysis uses several sources. First, the data on
GDP growth are from the Bureau of Economic Analysis (BEA). The BEA publishes changes
in GDP every quarter. This measure of real GDP is expressed as seasonally adjusted annual
rates in percentages, in 2005 dollar value. The choice for annual rates over quarterly growth
rates is for ease of comparison. In addition to this source, Datastream publishes GDP
estimates as well. More specifically, they submit seasonally adjusted flows fixed to 2005
dollars. Growth rates are obtained simply by computing the percentage change over the past
quarter. In order to compare this to the data of the BEA, annual rates are used.
Now let us turn to the key explanatory variable of the regression: SB activity. The data for
finance companies, funding corporations and asset-backed security issuers are found in the
Page 17
17
Flow of Funds from the Federal Reserve. These three intermediaries are used as a proxy for
SB activity similar to the paper by Adrian, Moench and Shin (2010). From the Flow of Funds,
financial assets and financial liabilities are collected for each intermediary. This is done for
flows and levels. The Federal Reserve puts both at the disposal of the public. In most cases,
the difference in levels equals the flows. However, this is not the case with the SB data
because of differences in market and historical values as well as statistical differences.
Therefore, we assess both types in our analysis. By deducting liabilities from total assets,
total equity is obtained. Then, financial leverage is computed as the ratio of financial assets
over equity. Again, growth rates of financial leverage are computed. One has to take into
account a possible change from negative to positive financial leverage. Therefore, absolute
values are used to compute a realistic growth rate. This is the measure used by Adrian,
Moench and Shin (2010) to account for shadow banking activity. However, it may be more
intuitive to use growth in assets or liabilities. Indeed, using this straightforward growth in
shadow banking activity may be a better proxy for the scope of SB. This is because the
computation of leverage in the case of small negative equities leads to extremely high
negative financial leverage. The asset or liability growth is simply calculated as the change in
assets or liabilities vis-à-vis the previous quarter.
We now have our main explanatory variable namely a proxy for SB activity. In the following
paragraph, the data for the control variables are reviewed. First, the rates for the certificates
of deposits are obtained from the Flow of Funds as well. Expressed in percentages per year,
these monthly data are averaged over three months to get to an average rate per quarter.
Similar data are obtained for the long-term interest rate, for which the thirty-year US Treasury
security yield is used. These yearly rates are given on a monthly basis and hence an
average per quarter is computed. For the yield curve, we use the 30-year interest rate minus
the three month yield on US Treasury securities. Inflationary data are based on the
Consumer Price Index, which contains quarterly growth rates for the USA. The CPI-index is
normalized to the period 1982-1984. The data are from the Datastream database. The
average quarterly oil price is obtained from the St. Louis Fed. These data are not seasonally
adjusted and are expressed in dollars per barrel. Of this, quarterly changes in the price are
computed. The money measure M2 is obtained from the Flow of Funds. The money supply
data are seasonally adjusted. They are given for every month, hence the average money
supply is calculated over a quarter and then the subsequent change over a quarter is
computed. Total Industrial Production gives a quarterly summation of industrial production
and capacity utilization of the major industries in the USA. Data are normalized to 2007. For
every quarter, growth rates are calculated. Unemployment data are obtained from the
Bureau of Labor Statistics (BLS). The unemployment rate is given as a monthly percentage
Page 18
18
of total population from 16 years to 65 years old. Again, quarterly data are computed by
taking the average over three months and the quarterly change is computed.
6 RESULTS
6.1 REGRESSION ANALYSIS
Using the GDP data from the Bureau of Economic Analysis as dependent variable, the
multivariate regression is run in Eviews. This basic regression uses all the suggested
variables (see the previous section) and yields a strong model. The results of the regression
are found in Table 2 below. For the rest of this paper, a variable is strongly significant at the
one percent level, normally significant at the five percent level and weakly significant at the
ten percent level. These levels of significance are denoted by ***, ** and * respectively. All
data are tested for autocorrelation by using a two-period lag LM test. Also, they are tested for
heteroskedasticity with a Breusch-Pagan-Godfrey test. In this case, the data are not
significantly autocorrelated however they show significant heteroskedasticity, at the 10
percent level. Therefore, we adjust the data using Newey-West standard errors to account
for this heteroskedasticity.
VARIABLE COEFFICIENT
C 0,0087
FLOWLEVCHANGE -0,0001*
THREEMONTH_COD -1,8133***
THIRTYTHREESPREAD -1,9351***
THIRTY_IR 2,1974***
IPCHANGE 1,1275***
INFLATION -0,2808**
MTWOCHANGE 0,3688*
OILPRICECHANGE -0,0011
UNEMPLOYRATECHANGE 0,0610
R-squared 52,58%
Adjusted R-squared 48,09% Table 2 Multivariate regression analysis with GDP change as dependent variable (GDP data are from BEA)
Table 2 shows that the model has strong explanatory power with an adjusted R-squared of
48 percent with seven significant variables out of nine. Contrary to our intuition, the
coefficient for change in leverage of SB levels is significantly negative, but only on the 10
percent level. To assess the pure effect of SB flow leverage on GDP, a univariate regression
is estimated. The results can be found in Table 3. In this case, the coefficient for the SB
variable is positive, however it is not significant with a p-value of 0,50. Moreover, the
explanatory power of this model is very low namely 0,34 percent. The findings for the
Page 19
19
coefficients of the SB variable in Table 2 and 3 are not what we expect. Indeed, as
mentioned before, our main hypothesis is that SB activity has a positive influence on the
growth of the economy. Is this a real result or is it due to the choice of the variable? These
questions are further examined below.
Now let us first turn to the coefficients of the other variables i.e. the control variables. The
short-term interest rate as measured by the three-month Certificate of Deposit rate is
negative in a strongly significant way, as expected. Indeed, a higher short-term interest rate
increases the cost of borrowing and will have a detrimental effect on GDP. Next, the
coefficient of the 30-year interest rate and the yield curve are significantly different from zero,
both on the 1 percent level. However, the signs of these variables are opposite to what is
expected. Indeed, a positive difference in the yield curve should predict economic growth
whereas a higher long-term interest rate forecasts a recession. A possible explanation for
this is the fact that the analysis uses only a one-period i.e. a one quarter lag. This might not
be enough in order to account for the long-term effects of the 30-year interest rate and the
shape of the yield curve, as investors might not have taken this into account. As noted
before, the coefficient for the industrial production should be positive. Our analysis confirms
this with a strongly significant coefficient of 1,13. Inflation is another variable that is
significant, at the 5 percent level. Moreover, the sign of the inflation coefficient is exactly as
expected. The same conclusion holds for the money supply variable, measured by M2. With
more money circulating in the economy, there is more room for investments. This is
confirmed by a positive and weakly significant coefficient. The coefficient of the change in oil
price is negative as expected, similar to inflation. However, this coefficient is not significant.
The last variable in the analysis is expected to be negative, as a higher unemployment rate
leads to less potential growth. This is not confirmed by the positive coefficient of 0,06, but
since it is not significantly different from zero, no conclusion can be made on this variable.
Let us now investigate whether the counterintuitive result of a negative SB leverage
coefficient is due to the choice of the independent variable, or to a legitimate effect. For this,
we replicate the regression analysis of Table 2 by replacing the variable that measures
change in flow leverage by other SB variables. The variables that are tested in the
multivariate regression are the following: the change in SB level leverage, the growth of SB
assets and the equity ratio. The SB level leverage is a gross measure whereas the SB flow
leverage is a net measure, as the latter only represents net transactions. The growth of total
SB assets might be a good measure because it is not biased as is the case with leverage.
This bias follows from the fact that small negative equities lead to very high negative
leverages, whereas large negative equities result in rather small negative leverages.
However, the latter should clearly give a worse result as large negative equities imply too
Page 20
20
many liabilities relative to assets. In order to avoid this bias, one can also use the equity
ratio, which equals the ratio of equity divided by assets.
Table 4, Table 5 and Table 6 can be found in the Appendix and display the results of the
multivariate regressions with these three different SB variables. The results of the different
regressions show that the model keeps its large explanatory power, independent of the
different SB variables. Indeed, the R-squared varies around 53 percent for all three models.
Moreover, the explanatory control variables remain significant, with the sign of the
coefficients as discussed previously. However, using different SB variables clearly gives
different results. Table 4 shows that the coefficient for the change in SB leverage using levels
is positive as expected, but is not significantly different from zero with a p-value of 0,20. A
similar conclusion holds for Table 5, which displays the results of the multivariate regression
using the SB asset growth as main independent variable. It is again positive but not
significant, and hence we cannot use this to assess the predictive power on real growth. The
last regression results in Table 6 show a negative coefficient for the equity ratio. This is again
as expected by intuition. Indeed, a higher equity ratio implies lower levered SB activities and
hence lower GDP growth. On the other hand, the coefficient is not significantly different from
zero. It is thus clear that the other SB variables are not useful in the model. Although they all
have the expected sign, no variable is significantly different from zero. In what follows,
robustness tests are performed using the SB variable from the first regression i.e. the change
in leverage of SB flows.
6.2 ROBUSTNESS TESTS
In this section, some robustness tests are performed. The outcomes of these robustness
tests can help one in interpreting the results of the basic regression. The tests in this section
are performed in particular to see whether the main variable i.e. the SB activity proxy
remains significantly different from zero. If the result is similar to the basic regression, we
might conclude that our SB proxy has real explanatory power on GDP, as it stays robust in a
different model. On the other hand, if the SB variable is no longer significant, the SB variable
is not robust. In this case, we cannot infer that SB activity is an important factor for
forecasting real growth.
As a first robustness test, a new model is set up whereby only the significant variables of
Table 2 are included. Again, GDP growth from the Bureau of Economic Analysis is used as
dependent variable. Table 7 summarizes the obtained results. One can see that the
significant variables are quite robust. Indeed, five out of seven variables have a significant
effect on GDP. Moreover, the R-squared is of the same size as that of the basic regression
(Table 2), even without two additional variables. However, our key variable namely the
Page 21
21
change in leverage of SB is not significant. The rest of the variables are of the same sign,
size and significance as in the base model except for the change in the money supply as this
is not significantly different from zero. The same comments hold as before. It is quite
counterintuitive that a larger long-term interest rate predicts a higher GDP. Also, the negative
effect of the term structure is not as expected. Both effects can be attributed to the one-
period lag, as mentioned in the previous section. This model can also be compared to the
basic model using the Akaike Information Criterion (AIC). The AIC for this model and the
basic model are -4,88 and -4,85 respectively. This implies that the new, less complex model
minimizes the information loss, albeit a small difference. We can thus conclude that this
model is better, as it is less complex but still has large explanatory power and significant
explanatory variables. Again, note that the SB variable is no longer significant with a p-value
of 0,47.
VARIABLE COEFFICIENT
C 0,0100
FLOWLEVCHANGE -0,0001
THREEMONTH_COD -1,7994***
THIRTYTHREESPREAD -1,9079***
THIRTY_IR 2,1736***
IPCHANGE 0,9470***
INFLATION -0,2760**
MTWOCHANGE 0,3650
R-squared 52,32%
Adjusted R-squared 48,88% Table 7 Regression results of less complex model using only significant variables of basic regression
A second test for robustness includes the same independent variables as in the original
model but we now change the definition for GDP growth. Instead of using the GDP growth
figures published by the BEA, one can use the seasonally adjusted GDP data from the
Datastream database, expressed in annual rates. The data are again adjusted using Newey-
West s.e. and the regression results can be found in Table 8. Similar results are obtained
using this GDP definition. The R-squared is around 52 percent as before and again seven
crucial variables are statistically significant. The change in the oil price and the change in the
unemployment rate are not significantly different from zero, which is the same conclusion as
in Table 2.
Page 22
22
VARIABLE COEFFICIENT
C 0,0092
FLOWLEVCHANGE -0,0001*
THREEMONTH_COD -1,8079***
THIRTYTHREESPREAD -1,9303***
THIRTY_IR 2,1824***
IPCHANGE 1,1044***
INFLATION -0,2800**
MTWOCHANGE 0,3570*
OILPRICECHANGE -0,0010
UNEMPLOYRATECHANGE 0,0563
R-squared 52,89%
Adjusted R-squared 48,43% Table 8 Basic regression using Datastream data for GDP
6.3 COMPLEMENTARY REGRESSIONS
This section is similar to the analysis of Adrian, Moench and Shin (2010). In their paper, they
perform some complementary regressions as well. These tests give further insight in the
robustness of the results. Two tests are performed: the first is a regression that only includes
data up until the crisis, the second is an in-sample test to assess the strength of the basic
model.
A critical question one might ask is whether the results are attributable to the financial crisis.
In order to investigate this, the data set is restricted until the second quarter of 2007, as the
financial crisis started in the summer of 2007. Table 9 gives the results for this sample. The
model holds quite well in terms of explanatory power as well as to the number of significant
variables. However, it is quite clear that the SB leverage variable does not continue to
perform well in the sense that it is no longer significant. Indeed, with a p-value of 0,71 it has
no longer a significant impact on GDP. Moreover, several other explanatory variables are
less significant or lost their significance entirely, compared to the basic regression results in
Table 2. For example, whereas the term spread variable is strongly significant in the basic
regression using all data, it is only significant at the 10 percent level up until 2007. Also,
inflation is no longer significantly different from zero, although its sign is still as expected. The
rest of the variables are also as discussed in the section on the results of the basic
regression.. Note that the explanatory power of this model is worse than when using all data.
The R-squared in this case is about 20 percentage points less than the basic regression
(Table 2). We might hence conclude that the results found in the latter are partially due to the
market movements in the period from 2007 until 2012.
Page 23
23
VARIABLE COEFFICIENT
C 0,0119
FLOWLEVCHANGE -0,0001
IPCHANGE 1,0410***
THIRTY_IR 1,4171**
THREEMONTH_COD -1,2315**
MTWOCHANGE 0,6012*
THIRTYTHREESPREAD -1,0510*
INFLATION -0,1915
OILPRICECHANGE -0,0101
UNEMPLOYRATECHANGE 0,0368
R-squared 36,90%
Adjusted R-squared 29,12% Table 9 Basic regression analysis using data up until the financial crisis
Next, we perform an in-sample test of the originally obtained results. Using the coefficients
from Table 2, one can fill in the actual values of the variables to obtain an estimated value for
GDP growth. To assess the in-sample predictive power of the model, one has to choose a
certain period to run the test against. The paper uses the period of 1992 to 2001 i.e. ten
years or 40 quarters. The reason for this choice is because it is the most recent data that is
not affected by the recent financial crisis and contains all necessary data points. As can be
seen from Figure 1, the estimated data fit the actual data well up until 1998. From then on,
the actual GDP growth is much more fluctuating. Moreover, the estimated direction often
differs from the actual trend. The grey line is the GDP series obtained from the Bureau of
Economic Analysis, whereas the black line represents the estimated data using the
coefficients of Table 2. It is clear that the modeled data are an underestimation of the
variance of the actual data, as the black line is less variable than the grey line.
CONCLUSION
This paper reviews the shadow banking system and assesses its impact on real economic
growth. There are several reasons why this analysis is useful. First, the literature on SB
activity is scarce. Only since 2000 there has been some research on SB. Moreover, these
papers mostly covered the definition, the scope and the process of SB. A second reason is
that the SB system has enormously increased in size: estimates for SB activities in the USA
are between 20 and 30 trillion USD. Our main hypothesis for this paper is that SB activity has
a positive impact on the real economy. Indeed, the SB system provides credit to institutional
investors. This in turn may have a positive effect on investments and hence on the growth of
Page 24
24
GDP. However, the drawbacks of SB are significant as well. The most important
disadvantages of SB are the increased complexity of the financial system and the higher
aggregate risk, which both played a large role in the recent financial crisis. In order to test our
hypothesis, a multivariate regression is used with GDP growth as the dependent variable.
The main explanatory variable is the change in leverage of SB flows. In addition to our main
explanatory variable, the analysis includes eight more control variables that predict GDP, as
proven by earlier literature. In spite of the strong explanatory power of the entire model, the
result of the regression is counterintuitive, as the coefficient for the SB variable is significantly
negative. When we test further by using different proxies for SB activity, this negative
influence disappears. However, we still do not find that SB has a significantly positive impact
on real growth. This neutrality of SB is confirmed by some robustness tests and
complementary regressions. Using the results of this paper, authorities might be inclined to
closely monitor and regulate the SB system, as the downsides are too large relative to the
gain in economic growth. Indeed, according to our analysis the gains are not significant. In
this case, regulators should restrict innovation and financial reform. In further research, it
might be interesting to use different proxies for SB activity. In order to accomplish this, more
and accurate data is needed on SB activity. This is the largest obstacle for future research.
Another point for research is an out-of-sample test using the estimated model of this paper.
This might give additional information of the explanatory power.
Page 25
25
APPENDIX
VARIABLE COEFFICIENT
C 0,0279***
FLOWLEVCHANGE 0,0001
R-squared 0,34% Table 3 The univariate effect of change in SB flow leverage on GDP (BEA data)
VARIABLE COEFFICIENT
C 0,0089
LEVELLEVCHANGE 0,0065
THREEMONTH_COD -1,8337***
THIRTYTHREESPREAD -1,9710***
THIRTY_IR 2,2050***
IPCHANGE 1,2020***
INFLATION -0,2673**
MTWOCHANGE 0,3972*
OILPRICECHANGE -0,0076
UNEMPLOYRATECHANGE 0,0841
R-squared 53,62%
Adjusted R-squared 49,23% Table 4 Effect on GDP (BEA) using multivariate regression with change in leverage of SB levels
VARIABLE COEFFICIENT
C 0,0085
SB_ASSETGROWTH 0,0307
THREEMONTH_COD -1,8440***
THIRTYTHREESPREAD -1,9203***
THIRTY_IR 2,2156***
IPCHANGE 1,1284***
INFLATION -0,2745**
MTWOCHANGE 0,3352
OILPRICECHANGE -0,0005
UNEMPLOYRATECHANGE 0,0684
R-squared 52,40%
Adjusted R-squared 47,89% Table 5 Effect on GDP (BEA) using multivariate regression with change in SB assets
Page 26
26
VARIABLE COEFFICIENT
C 0,0047
EQUITY_RATIO -0,0789
THREEMONTH_COD -1,8430***
THIRTYTHREESPREAD -1,9250***
THIRTY_IR 2,2801***
IPCHANGE 1,1031***
INFLATION -0,2672**
MTWOCHANGE 0,4054
OILPRICECHANGE -0,0008
UNEMPLOYRATECHANGE 0,0592
R-squared 52,44%
Adjusted R-squared 47,93% Table 6 Effect on GDP (BEA) using multivariate regression with equity ratio
Figure 1 Comparison of the actual GDP growth and estimated GDP growth in a sample of 1992-2001
-2%
0%
2%
4%
6%
8%
10%
ESTIMATED GDP GROWTH
ACTUAL GDP GROWTH
Page 27
27
REFERENCES
Adrian, T., & Ashcraft, A. B. (2012). Shadow Banking: A Review of the Literature. New York:
Federal Reserve Bank of New York Staff Report No.580.
Adrian, T., Covitz, D., & Liang, N. (February 2013). Financial Stability Monitoring. Federal
Reserve Bank of New York Staff Reports.
Adrian, T., Moench, E., & Shin, H. S. (2010). Financial Intermediation, Asset Prices and
Macroeconomic Dynamics. New York: Federal Reserve Bank of New York Staff Report No.
422.
Amery, S., & Van Deyck, P. (2012, June). The Shadow Banking System: Features,
Deficiencies and Solutions. 23p. KU Leuven: Faculty of Business and Economics.
Bakk-Simon, K. (2012, April). Shadow Banking in the Euro Area: An Overview. ECB
Occasional Paper Series No.133 , 38p.
Banerjee, A., Marcellino, M., & Masten, I. (2005). Leading indicators for Euro-area inflation
and GDP growth. Oxford Bulletin of Economics & Statistics 67(1) , pp. 785-813.
Camba-Mendez, G., Kapetanios, G., Smith, R. J., & Weale, M. R. (2001). Automatic Leading
Indicator of Economic Activity: Forecasting GDP Growth for European Countries. Royal
Economic Society 4(1) , 37p.
Dotsey, M. (1998). The Predictive Content of the Interest Rate Term Spread for Future
Economic Growth. Federal Reserve Bank of Richmond Economic Quarterly 84(3) , 21p.
El Khadraoui, S. (2012). Draft Report on Shadow Banking. European Parliament -
Committee on Economic and Monetary Affairs.
El Khadraoui, S. (2012). Shadow Banking Presentation. (p. 37p). KU Leuven: European
Parliament.
Estrella, A., & Mishkin, F. S. (1998). Predicting U.S. Recessions: Financial Variables as
Leading Indicators. The Review of Economics and Statistics 80(1) , pp. 45-61.
European Central Bank. (2013). Enhancing the Monitoring of Shadow Banking. ECB Monthly
Bulletin , 11p.
European Commission. (2012). Green Paper Shadow Banking. Brussels.
Federal Reserve Staff. (2012, April). Federal Reserve Presentation. Retrieved from Federal
Reserve: http://www.federalreserve.gov/newsevents/rr-commpublic/fr-staff-reg-reform-
presentation-20120405.pdf
Financial Stability Board. (2012, November). Global Shadow Banking Monitoring Report.
Retrieved from Financial Stability Board Website:
http://www.financialstabilityboard.org/publications/r_121118c.pdf
Page 28
28
Financial Stability Board. (2011, October). Shadow Banking: Strengthening Oversight and
Regulation. Retrieved from Financial Stability Board Website:
http://www.financialstabilityboard.org/publications/r_111027a.pdf
Kelleher, D., Hall, S., & Bradley, K. (September 2012). Cost of the Crisis.
http://bettermarkets.com/sites/default/files/Cost%20Of%20The%20Crisis_0.pdf: Better
Markets.
King, R., & Levine, R. (1993). Finance and Growth: Schumpeter Might Be Right. Quarterly
Journal of Economics , pp. 717-737.
Krieger, S. (2011). Maturity Transformation and Systemic Risk. Presentation (p. 17). New
York: Federal Reserve Bank of New York.
Marcellino, M. (2007, April). A comparison of time-series models for forecasting GDP growth
and inflation. Retrieved from European University Institute:
http://www.eui.eu/Personal/Marcellino/1.pdf
Pozsar, Z. (2008). The Rise and Fall of the Shadow Banking System. Regional Financial
Review , 14p.
Pozsar, Z., Adrian, T., Ashcraft, A., & Boesky, H. (2010). Shadow Banking. New York:
Federal Reserve Bank of New York Staff Report No. 458.