Introduction To Law and To Finance Unit 8 Interpreting Financial Sector Data

Introduction to Law and to Finance
Unit 8 Interpreting Financial Sector

Data
Contents
8.1 Introduction Cause and Effect
3
8.2 Is There a Correlation of Bank Rate and Bond Yield?
4
8.3 Do Countries Legal Systems Affect the Development of

Their Financial Systems?
14
8.4 Understanding the Use of Regression Analysis
15
8.5 Interpreting the Regression Results of La Porta,

Lopez-de-Silanes, Shleifer and Vishny
20
8.6 Conclusion
22
References and Websites
22
Unit Content
Unit 8 begins by explaining some of the more basic statistical tools which
financial experts and economists use to analyse relationships between
financial and economic variables. The unit demonstrates the use of charts
and simple correlation coefficients, and to provide a context, examines the
potential relationship between central bank interest rates and the yield on
government bonds. This part of the unit is structured like a research report,
to demonstrate the steps required to undertake research. The unit then
examines the more sophisticated method of regression, which allows analysts to examine the relationships between variables in more detail. This
method is widely used in financial and economic research. The unit explains
the essential components of regression analysis, and how to interpret the
results of estimation. The unit does not focus on the mechanics of regression
analysis; instead it explains what you need to know to understand regression results. To show how this is done, the unit interprets some research
results concerning the relation between legal origins, shareholder rights and
ownership concentration.
Learning Objectives
When you have completed your study of this unit and its readings, you will
be able to
formulate research questions and hypotheses concerning relations
between variables, financial and economic
consider the relevance and motivation for research
construct and interpret time charts of data
construct and interpret scatter plots of data
calculate and interpret correlation coefficients between two variables
explain the limitations of correlation
analyse causality between two variables
conduct and write research reports using these methods
identify the different types of data, time series and cross-section
discuss the essential elements of regression analysis
interpret estimated regression equations.
Reading for Unit 8

In Unit 8 you will read a short explanation from the Bank of England, on
how changes in the Banks official interest rate have effects on the level of
expenditure in the economy. This reading is included in the unit text, and
is also available online (see Reference section).
Course Reader
Rafael La Porta, Florencio Lopez-de-Silanes, Andrei Shleifer and
Robert Vishny (1998) Law and Finance.
University of London
Unit 8 Introduction to Law and Finance
8.1 Introduction Cause and Effect

Financial markets facilitate the buying and selling of large quantities of
financial assets such as bonds, equities, and currencies. The value of such
trading in normal periods greatly exceeds the recorded value of business in
important categories of goods and services. For example, the purchase and
sale of currency in foreign exchange transactions is much greater than the
purchase and sale of goods and services between countries.
On financial markets the balance of demand and supply for financial assets
determines their prices the price (and therefore yields) of bonds, the price
of shares, and the price (exchange rates) of currencies. Because those prices
can powerfully influence firms and individuals financial and economic
decisions, analysts use much effort and resources on trying to understand
what causes prices to be at a particular level or to change.
But how can we know what causes changes in asset prices or other economic
variables?
That specific question is an example of the type of question that underlies
the research that academic finance experts and economists engage with
continually. Their basic research questions have the general form of asking
Does X have a systematic causal effect on Y? Our question is How can we
can investigate whether X has an effect on asset prices or other economic
variables? where X stands for some measurable variable that, according to
prior reasoning, we think is likely to have such an effect.
In this unit we explain the most basic statistical concepts and tools that
analysts can use as elements in the wider task of studying causation. But
they come with a health warning: statistical tools enable us to uncover
relationships between financial and economic variables with great rigour,
but they alone do not demonstrate that a change in one variable causes a
change in another variable. A frequently stated wise maxim is:
Correlation is not causation.
Or, more generally,
Statistical association alone does not demonstrate cause and effect.
If researchers do succeed in understanding the causes of financial and
economic change, the strength of their conclusion stems from a complex
interaction between statistical techniques and abstract, theoretical reasoning.
Success also requires intelligent and painstaking measurement of data,
sound common sense, carefully targeted formulation of the research question, and several other virtues.
Nonetheless, in the course of this unit you will study only the basic statistical concepts instead of the full set of research methods. The objective of this
unit is to enable you to critically read research articles in your courses on
finance, not to enable you to carry out research in finance yourself. Many
academic publications on finance use advanced techniques that combine
statistical estimation with economic theory models or, in other words,
they use econometrics. But the publications you might encounter in finance
courses are at a level that can be understood without a knowledge of
Centre for Financial and Management Studies
advanced econometrics. The basic statistical concepts you study in this unit
will give you a good basis for understanding such publications.
Although statistical association alone does not demonstrate cause and effect,
statistical techniques that explore associations between phenomena are one
valuable element in exploring cause and effect. In this unit you study basic
techniques by examining their usefulness in answering two different types
of cause-effect questions.
The first question (studied in Section 8.2) is:
What effect does the Bank of Englands official interest rate have on
bond yields?
The second is a broader question about countries financial systems. It is one
that seems particularly appropriate for a programme on Finance and Financial Law. This question is studied in Sections 8.3 and 8.5.
Do countries different legal systems have an effect upon the extent to
which companies shares are concentrated in the hands of a few
owners or dispersed among many?
8.2 Is There a Correlation of Bank Rate and Bond Yield?

In Section 8.2 we consider how basic statistics can shed light on a question in
finance that has considerable practical importance. In the process, we
introduce two statistical tools:
scatter plots
correlation coefficient.
The unit is structured in the same way as a report on a simple research
project might be, in a series of logical steps:
1 first, set out the research question
2 describe the motivation for the research by outlining its real world
relevance and background
3 describe the data to be used
4 introduce the concepts of a time chart and a scatter plot, apply them to
the data, and discuss the results
5 introduce the concept of a correlation coefficient
6 calculate the correlation coefficient
7 discuss the results
8 finish with a summary of the conclusions, including
a) the answer to our research question suggested by our statistics
b) the limitations of the answer
c) suggestions for further research .
8.2.1 Research question the effect of bank rate

Bank Rate is a rate of interest set by the Bank of England. The yield on fiveyear British government bonds is determined by demand and supply in the
bond market. Our research question is:
To what extent does the level the Bank of England sets for Bank Rate
have a systematic effect upon the yield on five-year British
government bonds?
Note
Research projects with a rigorous approach should express the research question as a null
hypothesis and alternative hypothesis, so that the statistical exercise involves tests to see
whether the data rejects the null hypothesis. However, the research report frequently
omits describing hypotheses in those terms and, for an illustrative example like the
present one, it is adequate to work with a research question in the literary form that we
have used.
8.2.2 Motivation, background and relevance

Central banks have control over the interest rates they charge banks for
short-term credit, and one of these rates is used as a major instrument of
wider economic policy. In the United Kingdom the Bank of Englands policy
rate is Bank Rate. The Bank of Englands Monetary Policy Committee (MPC)
determines its level at its monthly Monetary Policy Committee meeting. Its
decisions are based upon its assessments of economic conditions and the
economic objectives of the Bank of England.
Experience in 2008 and 2009 provides an interesting example. With a view to
controlling inflation, Bank Rate was 5 per cent per annum (p.a.) in September 2008. At its meeting on 8 October 2008 the MPC, believing that the
economy was heading into recession following the crash of Lehmans Bank
in New York and that inflation would fall below its target level, cut Bank
Rate to 4.5 per cent. At its subsequent monthly meetings the MPC successively reduced Bank Rate until, on 5 March 2009, it cut it to 0.5 per cent in
order to contribute to stimulating economic recovery and to stabilising the
banking sector (while, formally, relating those wider objectives to its perspective on inflation, which was the Bank of Englands statutory remit).
How could the level of Bank Rate influence economic activity? Since its
direct influence is only upon the cost of the Bank of Englands short-term
credit to banks, the question relates to possible indirect channels.
Reading
Please now read the following extract where the Bank of England explains the standard
theory of those indirect channels. The reading is also available on line (see References).
When the Bank of England changes the official interest rate it is attempting to influence the
overall level of expenditure in the economy. When the amount of money spent grows more
quickly than the volume of output produced, inflation is the result. In this way, changes in
interest rates are used to control inflation.
The Bank of England sets an interest rate at which it lends to financial institutions. This
interest rate then affects the whole range of interest rates set by commercial banks, building
societies and other institutions for their own savers and borrowers. It also tends to affect the
price of financial assets, such as bonds and shares, and the exchange rate, which affect
consumer and business demand in a variety of ways. Lowering or raising interest rates affects
spending in the economy.
A reduction in interest rates makes saving less attractive and borrowing more attractive, which
stimulates spending. Lower interest rates can affect consumers and firms cash-flow a fall in
interest rates reduces the income from savings and the interest payments due on loans.
Borrowers tend to spend more of any extra money they have than lenders, so the net effect of
lower interest rates through this cash-flow channel is to encourage higher spending in
aggregate. The opposite occurs when interest rates are increased.
Lower interest rates can boost the prices of assets such as shares and houses. Higher house
prices enable existing home owners to extend their mortgages in order to finance higher
consumption. Higher share prices raise households wealth and can increase their willingness
to spend.
Changes in interest rates can also affect the exchange rate. An unexpected rise in the rate of
interest in the UK relative to overseas would give investors a higher return on UK assets
relative to their foreign-currency equivalents, tending to make sterling assets more attractive.
That should raise the value of sterling, reduce the price of imports, and reduce demand for UK
goods and services abroad. However, the impact of interest rates on the exchange rate is,
unfortunately, seldom that predictable.
Changes in spending feed through into output and, in turn, into employment. That can affect
wage costs by changing the relative balance of demand and supply for workers. But it also
influences wage bargainers expectations of inflation an important consideration for the
eventual settlement. The impact on output and wages feeds through to producers costs and
prices, and eventually consumer prices.
Some of these influences can work more quickly than others. And the overall effect of
monetary policy will be more rapid if it is credible. But, in general, there are time lags before
changes in interest rates affect spending and saving decisions, and longer still before they
affect consumer prices.
We cannot be precise about the size or timing of all these channels. But the maximum effect
on output is estimated to take up to about one year. And the maximum impact of a change in
interest rates on consumer price inflation takes up to about two years. So interest rates have
to be set based on judgments about what inflation might be the outlook over the coming
few years not what it is today.
Market
rates
Asset
prices
Ofcial
rate
Domestic
demand
Net external
demand
Expectations /
condence
Total
demand
Domestic
Inationary
pressure
Ination
Import
prices
Exchange
rate
Source: http://www.bankofengland.co.uk/images/from_int_inf2.gif
In this research we are concerned with the influence that Bank Rate has on
bond yields. As the Bank of England states, Bank Rate is believed to tend to
affect the price of financial assets, such as bonds. If a cut in Bank Rate raises
the market price of bonds, it lowers bond yields. For firms that consider
borrowing by issuing new bonds, that reduces the cost of borrowing and
could encourage them to borrow in order to finance investment in building
new production facilities. Therefore, knowledge of the extent to which
changes in Base Rate do influence bond yields enables us to understand a
potentially important channel from Bank Rate to economic activity.
Although the indirect channel through bond yields concerns yields on
corporate bonds, our research question is To what extent does the level the
Bank of England sets for Bank Rate have a systematic effect upon the yield on fiveyear British government bonds? Since we assume that corporate bond yields
equal government (risk-free) bond yields plus a spread or risk premium,
a systematic influence of Bank Rate on government bond yields implies a
similar influence on corporate bond yields, unless changes in Bank Rate
cause the risk premium to change.
In summary, then, an investigation into the influence of Bank Rate on
government bond yields is expected to provide information on the effect
that Bank Rate has on firms cost of borrowing in the bond market. If we
find, for example, that there is no such systematic influence, the cuts in Bank
Rate following the 2008 financial crisis are not likely to promote economic
expansion through that channel.
8.2.3 Data
We use data for Bank Rate and the Bank of Englands measure of the yield to
maturity of five-year government securities (see References for website).
Our data set consists of the values of Bank Rate and the five-year rate on the
last day of each month from 31 January 1982 to 28 February 2010.
8.2.4 Analysis of time chart and scatter plot

From those data we can obtain an initial conclusion about the relationship
by translating the numbers into their graphical equivalents.
A useful graphical tool with which to start is a time chart, showing the level
of Bank Rate and the level of the five-year rate at each point in time from 31
January 1982 to 28 February 2010. To be precise, since our data are monthly
observations, the time chart shows each monthly observation in that period
and joins the monthly observations with a line through the intermediate
days. By entering the data into Microsoft Excel (or similar spreadsheet
software) and using its chart functions we can produce a time chart easily.
Figure 8.1 shows the time chart we have created from our data series.
Let us look at the time chart. The horizontal axis marks the month of the
observations; we have 337 monthly observations of Bank Rate and the same
number of observations of the five-year government bond rate between 31
January 1982 and 28 February 2010. Note that there is not space on the
horizontal axis to list all of the months.
Figure 8.1 Time chart, Bank Rate and five-year government bond rate, Jan
1982Feb 2010
16
14
12
10
8
5-year government bond rate
6
4
2
29 Sep 08
29 Apr 07
29 Nov 05
31 Jul 04
29 Feb 03
29 Sep 01
29 Apr 00
29 Nov 98
31 Jul 97
29 Feb 96
29 Sep 94
29 Apr 93
29 Nov 91
01 Jul 90
01 Feb 89
01 Sep 87
01 Apr 86
01 Nov 84
Bank Rate
01 Jun 83
01Jan 82
The vertical axis measures levels of the interest rate and yield as per cent per
annum. The points on the dotted line show the level of Bank Rate at each
date. The points on the solid line show the level of the yield on five-year
government bonds on the same dates.
Review Question
Inspecting the chart, what light do you think it sheds on our research question?
Our first impression is that the chart shows that Bank Rate and the five-year
government bond rate are closely related. It appears that when Bank Rate is
being raised over a period, the five-year government bond rate generally
rises; and in periods when the Bank Rate is being reduced, the five-year
government bond rate also declines. In other words, our first impression
suggests a preliminary answer to the research question To what extent does
the level the Bank of England sets for Bank Rate have a systematic effect upon the
yield on five-year British government bonds? Our first impression is that there
does appear to be a systematic relationship, and this suggests there may be a
systematic effect from Bank Rate to the bond yield. Further investigation is
warranted!
Another graphical tool that enables us to view the data from a different
angle and draw conclusions is a scatter plot, as shown in Figure 8.2.
The horizontal axis in Figure 8.2 does not measure time, but, instead, measures Bank Rate as per cent per annum. The vertical axis measures the yield
on five-year government bonds as per cent per annum. The scatter plot is the
set of black dots. Each dot shows the level of the five-year government bond
yield (on the vertical axis) and the Bank Rate (on the horizontal axis) at one
observation date (one monthly date in our data series).
Figure 8.2 Scatter plot, Bank Rate and five-year Government Bond Rate, Jan
1982Feb 2010
5-year 16
government
bond rate
14
12
10
8
6
4
2
0 |
0
10
12
14
16
Bank Rate
Review Question
What can we deduce from the scatter plot? How can we interpret the chart?
The positions of the plots give us an indication of whether the bond yield is
generally high when Bank Rate is high, and low when it is low. To understand such an indication, let us look at a scatter plot of a different pair of
variables (sterling LIBOR an interbank interest rate for wholesale credit,
and Bank Rate) which, as shown in Figure 8.3, has a more extreme pattern.
Figure 8.3 Scatter plot, Bank Rate and LIBOR
LIBOR % 20
18
16
14
12
10
8
6
4
2
0 |
0
10
15
20
Bank Rate
In Figure 8.3 the pattern shows clearly that when one interest rate is high the
other is high, and a low level of one is matched by a low level of the other.
In Figure 8.2, which plots the variables that are the subject of our research
question, the relationship is less clear, because the dots do not so closely lie
upon a straight line. Our first impression from Figure 8.2 is that it shows a
clear relationship between Bank Rate and the bond yield. It is similar to and
in the same direction as the relationship between the variables in Figure 8.3.
But the relationship is less clear. For example, in Figure 8.2 in several places
the dots cluster in a vertical line indicating that when the Bank Rate is at the
level indicated by that line, the bond yield has been at several different
levels denoted by the height of the dots in that vertical line. Nonetheless,
Figure 8.2 seems to show that there is a broad positive relationship between
Bank Rate and the bond yield.
Does the relationship that we have found in the time chart and scatter plot
shed light on the answer to our research question? Before considering that, it
is necessary to make our judgment on the degree of association between the
Bank Rate and bond yield more precise.
8.2.5 Correlation coefficient

The correlation coefficient is a statistic, calculated from our data series,
which gives a precise measure of the degree of association between Bank
Rate and the bond yield.
The value calculated for a correlation coefficient between two variables may
be negative, zero, or positive. It may lie in the range minus one to plus one
(1 to +1). A correlation coefficient of 0 means that there is no statistical
relationship between the two data series. A correlation coefficient of +1
means that there is a perfect positive relationship, so that the dots in the
scatter plot lie exactly on an upward sloping line. A correlation coefficient of
1 means that there is a perfect negative relationship, so that the dots lie
exactly on a downward sloping line; higher levels of one variable are associated with lower levels of the other.
We can use Microsoft Excel (or similar software) to calculate the correlation
coefficient of our two data series, Bank Rate and five-year government bond
yields. Entering the data into an Excel worksheet and using the spreadsheets CORRELATION function we obtain a correlation coefficient of 0.91.
The correlation coefficient is written as r = 0.91. What does a correlation
coefficient of 0.91 signify?
Exercise
To explore the meaning of the correlation coefficient, please use a ruler or other hard
straight edge and, with a pencil, draw a straight line that looks as if it goes through (or
nearly through) the centre of the whole group of dots.
Using a different set of data as an example, Figure 8.4 shows the type of line you should
try to draw.
10
Figure 8.4 Example of scatter plot and fitting a line by hand

Y
If you have drawn a straight line through the centre of the whole group of
dots in Figure 8.2, the correlation coefficient of 0.91 means that the line will
have a positive slope.
It also tells us how close the dots are to the line. A scatter plot like that in
Figure 8.3 has a very high proportion of dots a high proportion of the
observations in the data set close to the centred straight line. In Figure 8.2 a
lower proportion of observations are close to the line, but the correlation
coefficients value of 0.91 indicates that slightly more than 80 per cent of the
variance in the pairs of Bank Rate and bond yield observed is represented by
that straight line. (The correlation coefficient squared, r2, indicates the
percentage of the variance in observations represented by the centred
straight line). In the context of much statistical research, 80 per cent is quite a
high percentage.
8.2.6 Discussion of results

It is tempting to believe that our time chart, scatter diagram, and calculation
of the correlation coefficient give the answer to our research question:
To what extent does the level the Bank of England sets for Bank Rate
have a systematic effect upon the yield on five-year British
government bonds?
Each tool shows that, for the period covered by our data set, when Bank
Rate is high the yield on five-year government bonds is also high, and when
Bank Rate is low we find that the bond yield is low. The correlation coefficient shows that that linkage accounts for about 80 per cent of the variance in
the pair-wise observations. Therefore, it seems that when the Bank of England raises or lowers the Bank Rate there is a systematic effect on the yield
on five-year government bonds, which changes in the same direction.
11
However, that conclusion is too strong. It is not a valid conclusion. The

charts and correlation coefficient tell us only that there is a strong association between Bank Rate and the bond yield. But the conclusion drawn in the
previous paragraph goes further. It assumes that the strong association arises
because the level of Bank Rate at one time has a systematic effect that causes
the bond yield to be at the level we observe at that time. But the data itself
or its charts and correlation coefficient give no basis for making that
assumption.
Why not assume that the strong relation we have discovered does show that
Bank Rate has a systematic effect on the bond yield? Because that assumption represents only one hypothesis and at least two alternative hypotheses
could also be consistent with the data. In other words, there could be two
other explanations for the association we have observed. Starting with the
hypothesis behind our research question, we have three alternative hypotheses which are consistent with our data:
1 Changes in Bank Rate cause changes in the market-determined yield
on five-year government bonds in the same direction.
2 Causation is reversed: it is changes in the market-determined yield on
five-year government bonds that cause the Bank of England to change
Bank Rate in the same direction.
3 Neither Bank Rate nor the bond yield has a direct effect on the other.
Instead, both are influenced in the same direction by some third factor,
which we have not included in our data set. For example, it may be
that both the Bank of Englands choice of Bank Rate and market
participants desired yield on government bonds are influenced in the
same direction by their expectations of price inflation in the economys
goods and services.
In order to interpret the implications of our charts and correlation coefficient
for our research question we could, as a first step, try to investigate whether
closer inspection of the data supports hypotheses 2 or 3. Lets look again at
Figure 8.1, which is reproduced here.
Figure 8.1 Time chart, Bank Rate and five-year government bond rate, Jan
1982Feb 2010
16
14
12
10
8
5-year government bond rate
6
4
2
12
29 Sep 08
29 Apr 07
29 Nov 05
31 Jul 04
29 Feb 03
29 Sep 01
29 Apr 00
29 Nov 98
31 Jul 97
29 Feb 96
29 Sep 94
29 Apr 93
29 Nov 91
01 Jul 90
01 Feb 89
01 Sep 87
01 Apr 86
01 Nov 84
01 Jun 83
01Jan 82
Bank Rate
Review Question
Is there any information that we could obtain from Figure 8.1 to help us decide between
Hypothesis 1 and Hypothesis 2?
We think that if Hypothesis 1 provides a good explanation of the association
between Bank Rate and bond yield we would expect to see changes in Bank
Rate precede changes in bond yield. But if Hypothesis 2 is valid we would
expect to see changes in Bank Rate follow changes in bond yield. What do
we see in Figure 8.1?
Visual inspection reveals several occasions when changes in Bank Rate
followed changes in market-determined bond yield. For example, from
September 1993 the bond yield rises (September 1993 is not listed on the
horizontal axis, but this is the month when the five-year bond yield begins
to rise, in this period in the middle of the chart), but the Bank Rate falls for
several following months and only rises after April 1994, seven months later.
As you can see by following the solid and dashed lines from those upturns,
increases in the bond yield continue to happen before increases in Bank
Rate, until the bond yield turns, begins to fall and is followed with a lag (or
time delay) by reductions in Bank Rate. A similar pattern of changes in Bank
Rate following changes in bond yield can be observed when bond yields
start to rise after month December 1998 and at several other periods.
Since it appears that in many months Bank Rate appears to follow the bond
yields of preceding months, the data seems to be consistent with Hypothesis
2, at least in part. It may also be consistent with Hypothesis 3 if we assume
that bond yields respond to expectations of inflation and the Monetary
Policy Committee does too, but only with a lag.
8.2.7
Conclusions
This research has examined the question of whether the Bank of Englands
policy instrument, Bank Rate, has an effect on a government bond yield. If it
does, it may have an effect on the yield on corporate bonds and thereby
affect firms cost of capital and investment decisions.
Using monthly data from 31 January 1982 to 28 February 2010 we find there
is a strong linear correlation between Bank Rate and the bond yield.
However, correlation is not the same as causation, and the correlation we
found is also consistent with two hypotheses that are alternatives to the
hypothesis that Bank Rate affects bond yields. Visual examination of time
lags suggests that the hypothesis that bond yields affect Bank Rate receives
additional support from the data. And the observed time lags could also be
consistent with the hypothesis that both rates respond to a third variable,
inflation expectations.
These results suggest that further investigation, using econometric techniques, is warranted.
13
8.3 Do Countries Legal Systems Affect the Development of

Their Financial Systems?
In this Section we introduce a basic econometric technique, Ordinary Least
Squares Regression, which is widely used for analysing economic and
financial data. We could explain its main features by applying the technique
to the research question examined in Section 8.2, but we think it would,
instead, be interesting to see how it has been used to analyse a broader
question.
In this section the research you study is a scholarly article published in 1998,
Law and Finance by Rafael La Porta, Florencio Lopez-de-Silanes, Andrei
Shleifer and Robert Vishny, which will demonstrate how to interpret research results.
This article has had a powerful effect, because it tackled questions that had
not been systematically investigated before, and its reasoning and conclusions initiated a large number of related studies with similar or critical
standpoints in the subsequent decade.
8.3.1 The legal origins, shareholder rights, and ownership concentration

idea
In their article La Porta, Lopez-de-Silanes, Shleifer, and Vishny (LLSV)
investigate several interesting hypotheses. Their central, hypotheses (which
are linked rather than alternative hypotheses) are:
1 Countries legal systems differ according to their origins (particularly
their importation from different imperial powers in earlier times).
2 Those differences generate different degrees of protection to company
shareholders.
3 In countries with systems that provide weak protection to
shareholders, that weakness causes the ownership of companies
shares to be highly concentrated in a few hands.
The authors propose ways to measure the relevant legal and other characteristics of countries and they analyse those statistics using Ordinary Least
Squares Regression and other techniques.
Their article is not easy to read, but the best way to understand their argument and their use of Ordinary Least Squares Regression is to plunge
straight in. We would like you to read the article now, but to make it manageable and useful we recommend that you read only some sections. When
you have completed the reading we use the article as a basis for explaining
how to assess the conclusions about cause and effect that can be based on
Ordinary Least Squares Regression studies.
Reading
Please turn to this article now, and study Section I, pp. 111317, Section II (including
Table 1) pp. 111726, Section VI (including Table 8) pp. 114551, and Section VII,
pp. 115152.
14
Rafael La Porta,
Florencio Lopez-deSilanes, Andrei Shleifer
and Robert Vishny
(1998) Law and
Finance, reprinted in
the Course Reader from
the Journal of Political
Economy.
Make sure your notes cover the main points carefully, noting particularly the type of
data used and the analytic techniques, which will be discussed further below.
8.4 Understanding the Use of Regression Analysis

The conclusions LLSV summarise in Section VII are partly based on the
results of Ordinary Least Squares Regression reported in their Table 8. When
you read empirical studies of finance whether academic studies or those
published by analysts, banks, central banks, and regulators you will often
encounter regression results presented in a similar way. They contain
numbers statistical measures of different types, and to make sense of
them we need to know what the numbers mean. Knowing what the numbers mean is similar to the requirement we had in Section 8.2 to understand
the meaning of a correlation coefficient in order to make sense of the data.
In this section, 8.4, we introduce the main concepts needed to interpret the
results of studies that use regression estimates.
They are:
random variables
regression equations
estimated regression coefficients
random errors
coefficient of determination
statistical significance of estimated regression coefficients.
8.4.1 Random variables

Financial variables such as interest rates fluctuate in a random manner.
Similarly, many other economic variables, many social indicators, and also
measures of variables in medicine, engineering and many other fields, all
fluctuate in a random manner. The task of econometric regression is to find
systematic elements in the random fluctuations of variables.
In Section 8.2 we were seeking a systematic relationship between two
variables, Bank Rate and the market-determined yield on five-year government bonds. Underlying the type of statistical techniques used there
(time charts, scatter plots and correlation coefficients) is the notion that the
combinations of Bank Rate and the bond yield that we observe fluctuate in a
random manner over time. The statistical research task is to try to identify a
systematic pattern within those randomly distributed observations; we
sought to identify a systematic association between Bank Rate and the bond
yield before proceeding to consider whether that association reflected a
cause and effect relationship.
The correlation coefficient is one statistic that measures the systematic
association. If the correlation coefficient is less than +1 (perfect positive
correlation) but greater than 1 (perfect negative correlation), it indicates
that the systematic association being measured does not account for all the
15
observed values of Bank Rate and the bond yield. But the association is
observed on average; the correlation coefficient measures the extent to
which observations in our data are close to a central (average) straight line
like the one you drew on the scatter plot. The observations that do not lie on
the line diverge from it by amounts that are assumed to be random.
Similar concepts of random variables underlie regression analysis. Before
examining them in that context, let us look more closely at the different
types of observations that are used in regression analyses.
Review Question
When you read the selected parts of the article by La Porta, Lopez-de-Silanes, Shleifer,
and Vishny, did you notice that the type of data they use is different from the type we
used in Section 8.2?
The main difference between the type of data used in the article and
the type used in our study of the relation between Bank rate and bond
yield is this. Our observations were taken in one economy (the United
Kingdom) at different points of time. The sequence of observations is
a set of time series data.
The observations that La Porta, Lopez-de-Silanes, Shleifer and Vishny use in
their regression estimates do not differ in time and are therefore not time
series data. Instead, for any one variable any one characteristic each
observation is drawn from one of 49 countries. The total sample of observations comprises a set of cross-section data.
In summary, time series data record differences over time, but cross-section
data record differences between entities (countries in this case) at a given
point in time, or time period. The idea that the observations are random
underlies statistical analysis using cross-section data in the same way as
analysis using time series data. Using cross-section data we use statistical
analysis to seek systematic relationships between variables across entities
(countries in this case) within the random pattern of observations of those
cross-country variables.
Apart from the time series/cross section difference between the two types of
data, you might have noticed others. An interesting difference is that the
analyses in Section 8.2 use data on variables, an interest rate and yield,
which are normally expressed numerically. By contrast, La Porta, Lopez-deSilanes, Shleifer and Vishny use data in the form of indexes they have
constructed to measure variables that are not usually expressed numerically.
For example, an important set of variables that they measure is the existence
or otherwise of a countrys laws to protect minority shareholders. The
construction and measurement of variables is an important aspect to be
examined in interpreting the results of any statistical research, but we shall
not examine it critically here.
16
8.4.2 Regression equations

Correlation, which you studied in Section 8.2, also calculated a systematic
relation between two variables that fluctuated randomly. So what is the
difference between the underlying ideas of correlation and regression?
Although regression analysis is related to correlation analysis, conceptually
these two types of analysis are very different. The main aim of correlation
analysis is to measure the degree of linear association between two variables, and this is summarised by a sample statistic, the correlation
coefficient. The two variables are treated symmetrically. Both are considered
random; there is no distinction between dependent and explanatory variables, and correlation itself carries no implication of causality in a particular
direction from one variable to the other. (Dependent and explanatory
variables are explained in the next paragraph.)
Regression analysis, however, can incorporate relationships between two or
more variables and the variables are not treated symmetrically. The dependent variable and explanatory variables are carefully distinguished. If we had
used regression analysis for the research question posed in Section 8.2, we
would have treated Bank Rate as the explanatory variable that we assumed
to be set independently. The bond yield would have been treated as the
dependent variable that takes random values. And regression analysis
would seek and estimate a systematic relation between the explanatory
variable and the dependent variable.
That example refers to two variables, a dependent variable and one explanatory variable. But regression analysis, unlike the correlation coefficient,
enables us to consider a greater number of variables, and we can estimate
the relationships between the dependent variable and more than one explanatory variable.
To illustrate that, please think back to the correlation coefficient we calculated in Section 8.2.5. The value we calculated for the correlation coefficient
implied that the systematic straight-line relation accounted for approximately 80 per cent of the variance in the observations of Bank Rate and the
bond yield. What of the 20 per cent not accounted for by that correlation?
Perhaps the deviations from the straight line of those observed combinations
of Bank Rate and bond yield were due to chance (random shocks). But
perhaps some of them are accounted for by a systematic relationship between market-determined bond yields and a variable that is distinct from
and not correlated with Bank Rate. In other words, suppose that there are
three elements in the variance of the yield on five-year UK government
bonds:
perhaps part (about 80 per cent) of the variance in the yield on fiveyear UK government bonds is accounted for by its systematic
relationship with UK Bank Rate
perhaps part (say 15 per cent) is accounted for by a systematic
relationship between the yield on UK government bonds and the
market yield on US Treasuries
and
perhaps the remaining 5 per cent is the result of random shocks
17
Our correlation coefficient cannot handle all those elements. But the statistics estimated from a regression equation can.
In its simplest generic form, a regression equation relating one dependent
variable (Y) to one explanatory variable (X1) is written:
Y = + 1 X1 + u
(8.1)
For the research question we studied in Section 8.2, Y, the dependent variable, would be the yield on five-year government bonds, and X1, the
explanatory variable, would be Bank Rate. We will explain , 1 and soon.
To take account of the possibility of a systematic relationship between the
dependent variable and a second explanatory variable, we can write the
regression equation as
Y = + 1 X1 + 2 X 2 + u
(8.2)
In the example we have outlined, the second explanatory variable, X2, would
be the market yield on US Treasuries.
8.4.3 Estimated regression coefficients

The process of using our data set to estimate a regression equation like
equation (8.1) generates values for and 1. They are estimated regression
coefficients. To understand them, we concentrate on 1 alone. In equation
(8.1), the estimated value of 1 tells us how much Y would change in response to a change in X1 if the only element of change in Y is caused by its
systematic relationship with X1. Y would change by the change in X1 multiplied by 1.
Now let us consider the estimated regression coefficients for an equation like
equation (8.2) where we have two explanatory variables. The regression
produces estimated regression coefficients as values for , 1 and 2. Now
each estimated regression coefficient takes account of the fact that the
dependent variable is systematically related to more than one explanatory
variable. Consequently, the meaning of the estimated 1 is now different
from the interpretation in equation (8.1).
For equation (8.2) the estimated value of 1 tells us how much Y would
change in response to a change in X1, taking account of the fact that X2 has an
effect, but assuming that X2 is held constant so that the contribution of X1 is
isolated. Similarly, the estimated value of 2 tells us how much Y would
change in response to a change in X2, taking account of the fact that X1 has an
effect, but assuming that X1 is held constant so that the contribution of X2 is
isolated.
8.4.4 Random errors

The estimated regression coefficients measure the systematic relations
between the dependent variable and explanatory variables. But in econometrics we assume that systematic relations do not account for all the
variance in the dependent variable. A final element is random shocks, which
produce deviations of Y from the values it would have if it were completely
18
determined by its systematic relations with explanatory variables. Those

deviations are known as random errors or random disturbances.
The random errors are represented by u in equations (8.1) and (8.2).
Our regression estimates are good, unbiased estimates only if the random
errors fit a distribution with a mean of zero and if the distribution has other
important qualities. Consequently, any academic paper reporting econometric research includes several statistical tests to determine the properties of
the distribution of the random errors.
8.4.5 Coefficient of determination

One statistic produced in a regression is the coefficient of determination,
which is written as R2. (Usually it is adjusted to take account of the sample
size and the number of explanatory variables, and is written as Adjusted R2).
The coefficient of determination gives us information similar to that provided by the correlation coefficient in a correlation study. In Section 8.2 you
saw that the correlation coefficient squared (r2) measures the proportion of
the observations variance accounted for by their systematic association in a
linear correlation. Similarly, the coefficient of determination in a regression
equation tells us the proportion of the observed variance in Y that is accounted for by the estimated systematic relationships with the explanatory
variables.
8.4.6 Statistical significance of estimated regression coefficients

An important and necessary part of any regression is a test for the statistical
significance of the estimated regression coefficients. The regression gives
estimated values for coefficients such as 1 and 2 and, at first sight, if we fail
to carry out tests of statistical significance, we might deduce that the estimated coefficients show that systematic relationships between the
dependent variable and the explanatory variables do exist. But the tests of
statistical significance are a basis for a sounder judgment.
Their relevance stems from the following. Our estimates are derived from
the sample of observations contained in our data set. However, if our
research question is whether X1 and X2 systematically cause changes in Y, we
are referring to a general relationship that goes beyond the particular sample
of observations in our data set. In relation to the study we conducted in
Section 8.2, we would need to know whether a systematic relationship exists
beyond the monthly observations from 31 January 1982 to 28 February 2010,
although we are only using that sample of observations for our estimate. In
relation to the study by La Porta, Lopez-de-Silanes, Shleifer and Vishny, we
need to know whether the estimated regression coefficients calculated with
data from 49 countries at one time would apply if we used a larger sample.
The tests for statistical significance use the properties of probability distributions to calculate the degree of confidence we can have that the estimated
values of the regression coefficients estimated with our sample of observations are valid for the whole population of possible observations.
How do we test for the statistical significance of estimated regression
coefficients? An estimated regression coefficient is the mean of the relation-
19
ship, in the data sample, between its explanatory variable and the dependent variable (technically, it is the conditional mean). We can also estimate the
standard error of that estimated mean value.
A simple test for statistical significance is to use that estimated value and the
estimated standard error to calculate the degree of confidence with which
we can accept the estimated value as the value that would apply in the
whole population of possible observations. A standard criterion is to require
a 95 per cent confidence interval; if it is met it can be stated that the estimated regression coefficient is significant at the 5 per cent level. That
criterion is generally satisfied if the estimated value of the regression coefficient is at least twice as large as its standard error.
Using the test we have described, a quick inspection of an estimated regression equation enables us to determine whether the estimated regression
coefficients, calculated with our sample of data, do show that there is a
generally valid systematic relation between the explanatory and dependent
variable. It is common practice to reject as statistically insignificant any
coefficient that has a standard error greater than half the value of the estimated coefficient. In that case we would say that at that confidence level
there is no statistically significant evidence of a systematic relationship
between the relevant explanatory variable and the dependent variable.
However, an estimate which does not meet the 5 per cent significance level
might meet the (lower criterion of the) 10 per cent significance level, and
might be accepted on that basis.
8.5 Interpreting the Regression Results of La Porta, Lopezde-Silanes, Shleifer and Vishny
Let us now use the concepts explained in Section 8.4 to see how La Porta,
Lopez-de-Silanes, Shleifer and Vishny interpret their regression estimates.
Their results are set out in their Table 8, which is reproduced in Table 8.1
below.
The subtitle tells us that the dependent variable, Y, is the countries measure
of the mean ownership in listed companies by the three largest shareholders;
it is the authors measure of the degree of concentration in shareholder
ownership of companies.
The left-hand column lists explanatory variables (independent variables).
The earlier parts of the article set out the authors hypotheses about the
potential impact of each of the possible explanatory variables.
The middle column shows the estimated coefficients of a cross-section
regression when the regression equation includes some of the full list of
explanatory variables but omits others. It also shows the standard error of
each estimated regression coefficient in parentheses beneath the coefficient.
The right-hand column shows the estimated coefficients of a cross-section
regression when the regression equation includes all of the full list of explanatory variables. It also shows the standard error of each estimated
regression coefficient in parentheses beneath the coefficient.
20
Let us look at the right-hand column carefully. It shows the estimated

regression coefficients that have been calculated for each explanatory
variable. However, it also shows that only six of them have standard errors
which are small enough (relative to the estimated coefficient) to enable us to
accept them as meeting the criterion of significance at the 10 per cent level
(or, better, significance at the 5 per cent level; or the even tougher criterion of
significance at the 1 per cent level). The authors can reject as not statistically
significant the explanatory variables that do not meet those criteria; for their
coefficients the asterisks are omitted.
Table 8.1
Regression Results from LLSV
The Adjusted R2 is shown at the bottom of the column. It indicates that the
estimated systematic relationships of the dependent variable with the full
list of explanatory variables in the right-hand column accounts for 73 per
cent of the variance in the dependent variable. Such a result is often expressed by writing that the explanatory variables explain 73 per cent of the
variance, but that expression is potentially misleading.
21
Exercise
To consolidate your understanding of the regression conducted by La Porta, Lopez-deSilanes, Shleifer and Vishny, please try the following two tasks:
From the right-hand column of their Table 8, please identify and name the
explanatory variables that were found to have no statistically significant relation
with the mean ownership variable, and those which were found to have a
statistically significant relation.
From your reading of the LLSV article, please consider the reasons why the variables
which were found to have a statistically significant relationship with the mean
ownership variable were expected to do so.
8.6 Conclusion
Units 6, 7 and 8 have provided an introduction to some of the theory and
methods used in finance. In Unit 6 you learned about the yield or return on
bonds, and you examined the inverse relation between the yield on a bond
and the bond price. You also learned about the return on shares. In that unit
you considered the yield on bonds as the price of time. In Unit 7 you considered that investors also require a return for risk. In particular, you
examined how financial risk is measured, including volatility risk (risk of
price fluctuations, up and down) and downside risk (risk of loss). There you
examined probability distributions, and studied how to interpret the mean
as a measure of expected value and the standard deviation as a measure of
volatility. In Unit 8 you learned how to use some basic statistical tools to
conduct research on the possible relations between variables, using time
charts, scatter plots, and the correlation coefficient. In this unit you also
studied how to interpret regression analysis, a powerful and widely used
method used to estimate the possible relationships between variables.
We hope you have enjoyed and benefited from your study of these units.
The theory and methods should enable you to understand and interpret
financial reports and articles in your work, and/or reports and articles you
may read in your courses on finance. In addition, these units should be
useful preparation if you wish to develop further your understanding of
financial and quantitative techniques.
References and Websites

Bank of England statistics: http://www.bankofengland.co.uk/statistics/
Bank of England, How Monetary Policy Works,
http://www.bankofengland.co.uk/monetarypolicy/how.htm,
La Porta R, F Lopez-de-Silanes, A Shleifer and R Vishny (1998) Law and
Finance, Journal of Political Economy, Vol. 106, No. 6, December, pp, 11331155.
22

Introduction To Law and To Finance Unit 8 Interpreting Financial Sector Data

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Introduction To Law and To Finance Unit 8 Interpreting Financial Sector Data

Diunggah oleh

Hak Cipta:

Format Tersedia

Introduction to Law and to Finance

Unit 8 Interpreting Financial Sector

8.2 Is There a Correlation of Bank Rate and Bond Yield?

8.3 Do Countries Legal Systems Affect the Development of

8.4 Understanding the Use of Regression Analysis

8.5 Interpreting the Regression Results of La Porta,

References and Websites