Patience and "good" institutions: Predicting Real Interest Rates and Institutional Quality?
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Patience and "good" institutions: Predicting Real Interest Rates and
Institutional Quality?*
Atin Basuchoudhary**
Department of Economics and Business
Virginia Military Institute
Lexington, VA 24450
Phone: 540 464 7450
Email: basuchoudharya@vmi.edu
John David
Department of Applied Mathematics
Virginia Military Institute
Lexington, VA 24450
Chap Michie
Department of Applied Mathematics
Virginia Military Institute
Lexington, VA 24450
Abstract: The folk theorem suggests economic institutions that support coordination in markets
based in specialization should be quite common as long as people are patient. We test and find
supporting evidence for this hypothesis using non-parametric data driven approaches. These
approaches suggest that institutions are quantifiably good predictors of real interest rates, even after
controlling for the vagaries of monetary policy. This suggests that real interest rates may potentially
serve as a proxy for good institutions.
*Very rough draft. Please do not quote without permission.
**Corresponding author.
1|PageIntroduction.
The public choice literature has long acknowledged the tension between predation and beneficial
trade in the context of classic coordination games (Mueller, 1989, pp. 9 - 25). This literature looks
towards a social contract, the state as it were, that moves individuals from the predatory to the
mutually beneficial Pareto improving outcome (Buchanan, 1975). Moreover, this social contract
itself is a public good and a matter for public choice. Nevertheless, the public choice literature
analyzing the role of the evolution of this social contract has hitherto focused on the size of the
community, reliance on formal sanctions, and police enforcement of this social contract (Mueller,
1989). On the other hand, Folk theorems for infinitely repeated games suggest that that all feasible
payoffs that payoff dominate the individually rational payoff vector can be supported as payoffs of
sub-game perfect equilibria using trigger strategies as long as discount rates are sufficiently close to 1
(Fudenberg & Maskin, 1990). This suggests that cooperation should be quite common as long as
people are patient. In other words, the sort of cooperation implied by a social contract that reduces
free riding and predation and moves society towards a productive Pareto improving equilibrium is
more likely when people are patient, i.e. interest rates are low. In fact, a recent paper
(Basuchoudhary & Razzolini, 2013), outlines a formal theory relating the evolution of cooperation
and patience.
However, it has been hard to find evidence of cooperation among humans when
experimental designs simulate institutions by manipulating patience through game termination rules (
(Dal-Bo & Frechette, 2011). Part of this problem may be attributed to faulty experimental design
that does not take into account patience endogeneity (Basuchoudhary, et al., 2012; Basuchoudhary,
et al., 2010). Of course, evidence for cooperation abounds in the natural world (Nowak & Highfield,
2011). Popular writers ranging from Diamond (2005) to Wright (2001) trace the evolution of
cooperation as an institutional matter that follows the arc of history. Greg Clark (2007, p. 167, and
2|Page2005), points to coordination between technology, institutions, and people as the proximate cause
for the end of Malthusian economies. In fact, he notes that interest rates fell quite precipitously in
England during this transition from a Malthusian to a modern economy (Table 9, Clark, 2007). The
literature in economic growth also suggests that patience is critical for growing economies (Romer,
2001). All of this suggests that the Folk Theorem is indeed at work in generating wealth. Therefore,
as a matter of empirical fact, low real interest rates ought to be associated with good institutions. We
investigate this link in this paper.
Interest rates are a consequence of how people feel about the future and as such arise out of
some psychological process that links the availability of money and interactions between people, the
institutions that envelop their lives and the evolution of the these very institutions. Then, any
attempt to link real interest rates to institutions is fraught with endogeneity problems. Thus, any
parametric approach is open to criticism from two fronts. First, there is the risk that a regression
based approach misses a variable that explains both institutional evolution and real interest rates.
Second, even if good instruments are available to address the first problem, economists tend to use
regression based tools to increase in sample fit rather than predictive power. Our non-parametric
approach eliminates both problems and can predict real interest rates much better than a simple
average. Thus, our approach provides an empirical link between real interest rates and institutions
that eliminates many of the theoretical problems that arise from investigating a dataset that may not
be distributed in way that adhere to all the assumptions required for a scientifically plausible
statistical investigation (Einav & Levin, 2013). Moreover, our ability to use institutional variables to
predict real interest rates provides for the first time some real evidence on the role of interest rates
as an important factor that links good institutions and the evolution of Pareto improving social
contracts. Further, if institutional variables are good predictors of real interest rates then interest
rates could potentially serve as a proxy for institutional variables in an observable way.
3|PageIn what follows we try to predict real interest rates using money growth and indices of
institutions thought to matter for economic growth. We suggest that if institutional variables can
predict real interest rates then institutions generate patience – possibly by reducing uncertainty about
the future (cite North). We run a horse race among many prediction techniques and find that
ensemble based decision tree analysis provides the most accurate predictions. We then use that
technique to find the institutional variables that matter the most for predicting interest rates. Thus,
our paper introduces two innovations. First, we find some empirical evidence for the theoretical link
between institutions and interest rates. Second, we use atheoretical data mining techniques to
investigate whether institutional data can predict real interest rates.1
We describe our data in Section 1. We spend a considerable amount of time describing our
scientific methodology in Section 2. Results are described in Section 3. Section 4 concludes.
Section 1. Data.
We get the data for real interest rates from the World Bank and our institutional variables from the
International Country Risk Guide (ICGR).2 Our dataset includes 37 countries(reported in Table 1 in
the Appendix) for which we have complete annual data from 1984 to 2007.
A set of 22 components, categorized as political, financial, and economic factors proxy
institutional health. These components were chosen to give an accurate assessment of the different
atmospheres of first and third world nations. Every component within the model is assigned a
numeric value, with the lower values indicating high risk and higher values meaning lower risk. This
data is then adapted into risk points for all the factors on the basis of a consistent pattern of
evaluation. The political risk assessments used in my thesis are produced through subjective analysis
1 We note here that at Susan Athey, a recent winner of the John Bates Clark medal in economics, made an
impassioned ple for using data mining techniques to improve the predictive power of economic analysis at a
plenary session at the 2013 Southern Economic Association meetings in Tampa, Florida.
2 International Country Risk Guide Methodology
4|Pageof the existing information. ICRG also provides the information and data on which the ratings for
the individual risk components are determined, together with its interpretation of that information
or data. This allows users of the model to weigh their own interpretation of the information and
data against the ICRG staff.
Table 1. Variable Descriptions3
Variable Name Description Point
Range
(Max)
Real Interest Rate (RIR) An interest rate that has been adjusted to remove the N/A. The
effects of inflation to reflect the real cost of funds to informati
the borrower, and the real yield to the lender. The on is
real interest rate of an investment is calculated as the provided
amount by which the nominal interest rate is higher by World
than the inflation rate. Bank. 4
Real Interest Rate = Nominal Interest Rate -
Inflation (Expected or Actual)
The real interest rate is the growth rate of purchasing
power derived from an investment. By adjusting the
nominal interest rate to compensate for inflation, you
are keeping the purchasing power of a given level of
capital constant over time.
Government Stability This is an assessment both of the government’s 12
ability to carry out its declared program(s), and its
ability to stay in office. The risk rating assigned is the
sum of three subcomponents, each with a maximum
score of four points and a minimum score of 0
points. A score of 4 points equates to Very Low Risk
and a score of 0 points to Very High Risk.
The subcomponents (of equal weight) are
● Government Unity
● Legislative Strength
● Popular Support
Socioeconomic Conditions This is an assessment of the socioeconomic pressures 12
at work in society that could constrain government
3 Descriptions are from International Country Risk Guide Methodology at
http://www.prsgroup.com/ICRG_methodology.aspx
4 http://data.worldbank.org/indicator/FR.INR.RINR
5|Pageaction or fuel social dissatisfaction. The risk rating
assigned is the sum of three subcomponents, each
with a maximum score of four points and a
minimum score of 0 points. A score of 4 points
equates to Very Low Risk and a score of 0 points to
Very High Risk.
The subcomponents (of equal weight) are
● Unemployment
● Consumer Confidence
● Poverty
Investment Profile This is an assessment of factors affecting the risk to 12
investment that are not covered by other political,
economic and financial risk components. The risk
rating assigned is the sum of three subcomponents,
each with a maximum score of four points and a
minimum score of 0 points. A score of 4 points
equates to Very Low Risk and a score of 0 points to
Very High Risk.
The subcomponents (of equal weight) are
● Contract Viability/Expropriation
● Profits Repatriation
● Payment Delays
Internal Conflict This is an assessment of political violence in the 12
country and its actual or potential impact on
governance. The highest rating is given to those
countries where there is no armed or civil opposition
to the government and the government does not
indulge in arbitrary violence, direct or indirect,
against its own people. The lowest rating is given to a
country embroiled in an on-going civil war. The risk
rating assigned is the sum of three subcomponents,
each with a maximum score of four points and a
minimum score of 0 points. A score of 4 points
equates to Very Low Risk and a score of 0 points to
Very High Risk.
The subcomponents are:
Civil War/Coup Threat
Terrorism/Political Violence
Civil Disorder
6|PageExternal Conflict The external conflict measure is an assessment both 12
of the risk to the incumbent government from
foreign action, ranging from non-violent external
pressure (diplomatic pressures, withholding of aid,
trade restrictions, territorial disputes, sanctions, etc)
to violent external pressure (cross-border conflicts to
all-out war).
External conflicts can adversely affect foreign
business in many ways, ranging from restrictions on
operations to trade and investment sanctions, to
distortions in the allocation of economic resources,
to violent change in the structure of society.
The risk rating assigned is the sum of three
subcomponents, each with a maximum score of four
points and a minimum score of 0 points. A score of
4 points equates to Very Low Risk and a score of 0
points to Very High Risk.
The subcomponents are:
War
Cross-Border Conflict
Foreign Pressures
Corruption This is an assessment of corruption within the 6
political system. Such corruption is a threat to
foreign investment for several reasons: it distorts the
economic and financial environment; it reduces the
efficiency of government and business by enabling
people to assume positions of power through
patronage rather than ability; and, last but not least,
introduces an inherent instability into the political
process.
The most common form of corruption met directly
by business is financial corruption in the form of
demands for special payments and bribes connected
with import and export licenses, exchange controls,
tax assessments, police protection, or loans. Such
corruption can make it difficult to conduct business
effectively, and in some cases may force the
withdrawal or withholding of an investment.
Although our measure takes such corruption into
account, it is more concerned with actual or potential
corruption in the form of excessive patronage,
nepotism, job reservations, 'favor-for-favors', secret
party funding, and suspiciously close ties between
politics and business. In our view these insidious
sorts of corruption are potentially of much greater
7|Pagerisk to foreign business in that they can lead to
popular discontent, unrealistic and inefficient
controls on the state economy, and encourage the
development of the black market.
The greatest risk in such corruption is that at some
time it will become so overweening, or some major
scandal will be suddenly revealed, as to provoke a
popular backlash, resulting in a fall or overthrow of
the government, a major reorganizing or
restructuring of the country's political institutions, or,
at worst, a breakdown in law and order, rendering
the country ungovernable.
Military in Politics Higher ratings The military is not elected by anyone. 6
Therefore, its involvement in politics, even at a
peripheral level, is a diminution of democratic
accountability. However, it also has other significant
implications.
The military might, for example, become involved in
government because of an actual or created internal
or external threat. Such a situation would imply the
distortion of government policy in order to meet this
threat, for example by increasing the defense budget
at the expense of other budget allocations.
In some countries, the threat of military take-over
can force an elected government to change policy or
cause its replacement by another government more
amenable to the military’s wishes. A military takeover
or threat of a takeover may also represent a high risk
if it is an indication that the government is unable to
function effectively and that the country therefore
has an uneasy environment for foreign businesses.
A full-scale military regime poses the greatest risk. In
the short term a military regime may provide a new
stability and thus reduce business risks. However, in
the longer term the risk will almost certainly rise,
partly because the system of governance will be
become corrupt and partly because the continuation
of such a government is likely to create an armed
opposition.
In some cases, military participation in government
may be a symptom rather than a cause of underlying
difficulties. Overall, lower risk ratings indicate a
greater degree of military participation in politics and
a higher level of political risk.
8|PageReligious Tensions Religious tensions may stem from the domination of 6
society and/or governance by a single religious group
that seeks to replace civil law by religious law and to
exclude other religions from the political and/or
social process; the desire of a single religious group
to dominate governance; the suppression of religious
freedom; the desire of a religious group to express its
own identity, separate from the country as a whole.
The risk involved in these situations range from
inexperienced people imposing inappropriate policies
through civil dissent to civil war.
Law and Order Law and Order are assessed separately, with each 6
sub-component comprising zero to three points. The
Law sub-component is an assessment of the strength
and impartiality of the legal system, while the Order
sub-component is an assessment of popular
observance of the law. Thus, a country can enjoy a
high rating – 3 – in terms of its judicial system, but a
low rating – 1 – if it suffers from a very high crime
rate of if the law is routinely ignored without
effective sanction (for example, widespread illegal
strikes).
Ethnic Tensions This component is an assessment of the degree of 6
tension within a country attributable to racial,
nationality, or language divisions. Lower ratings are
given to countries where racial and nationality
tensions are high because opposing groups are
intolerant and unwilling to compromise. Higher
ratings are given to countries where tensions are
minimal, even though such differences may still exist.
Democratic Accountability This is a measure of how responsive government is 6
to its people, on the basis that the less responsive it
is, the more likely it is that the government will fall,
peacefully in a democratic society, but possibly
violently in a non-democratic one.
Bureaucracy Quality The institutional strength and quality of the 4
bureaucracy is another shock absorber that tends to
minimize revisions of policy when governments
change. Therefore, high points are given to countries
where the bureaucracy has the strength and expertise
to govern without drastic changes in policy or
interruptions in government services. In these low-
risk countries, the bureaucracy tends to be somewhat
9|Pageautonomous from political pressure and to have an
established mechanism for recruitment and training.
Countries that lack the cushioning effect of a strong
bureaucracy receive low points because a change in
government tends to be traumatic in terms of policy
formulation and day-to-day administrative functions.
Section 2. Computational Techniques.
Data Mining is a large area in science, engineering and mathematics. It is receiving an increasingly
large amount of attention in both the technical and general literature, particularly as the amount of
available data has rapidly increased. In fact, it was the 2012 topic of Math Awareness Month. While
there are many techniques in this field and an abundance of literature on the subject, we will focus
on two techniques: artificial neural networks (ANNs) and decision trees (DTs).
An ANN is a mathematical model inspired by the way the human brain processes
information. It can also be thought of as a generalized nonlinear multivariate regression. See Hand
(2001), Alpaydin (2010), Kononenko (2007) and Bishop (2006) for an introduction to neural
networks and data mining in general. The fundamental building block of an ANN is the artificial
neuron. An artificial neuron is simply a composition of mathematical functions operations.
Assuming the input from the previous layer is denoted by a vector, , the
artificial neuron outputs, , have the form,
∑
where g is the generally nonlinear transfer function, , are the weights and are the biases. An
ANN consists of an interconnected collection of artificial neurons. For example a two layer ANN
with linear transfer functions in the second layer would take on the form,
10 | P a g e∑ (∑ )
where the superscript indicates the layer in which the weight is used and denotes the transfer
function for the kth neuron.
The data is split into 3 categories when finding the optimal weights for these networks, i.e.
training the networks. The training data is the portion of the data in which the optimization
algorithm uses to find the weights. The validation data is the portion of the data that is used to
determine when to stop training to avoid overfitting. The testing data is the portion of the data that
has no effect on training and is used to validate the model.
There is a large number of decisions one has to make when using these networks, including
the number of layers, number of neurons, how to portion the data into the various classes,
optimization routine and types of transfer functions. We experimented with a large number of these
options but ultimately settled on using feedforward backpropagation newtworks, with 2 hidden
layers, 8 neurons in the first layer and 6 in the second, tan-sigmoid transfer functions in the first two
layers and linear in the third, the Levenberg-Marquardt training routine was used and the data was
partitioned into training, validation and testing randomly at percents 70%, 15%, and 15%
respectively. The Matlab ANN toolbox was used to perform the computations in this work.
In general, a decision tree (DT) is a graph which takes options and weighs them by
observations to most properly group them. A decision tree groups these final outcomes by
classification, where a regression tree groups them by specific numbers. A very simple regression
tree would be one to help you decide how much to tip. For example, if your service was very fast
(waited less than 15 minutes for your food) and the waiter kept your glass filled (more than 2 refills)
you may want to give a good 20 % tip. For example if your service was slow (waited more than 15
11 | P a g eminutes for your food) and your drink was empty often (less than 2 refills) you may want to give a
low 10 % tip. All the possible results for this regression tree can be seen in Figure 1.
Figure 1
Figure 1. A simple example of a DTT designed to predict tip amounts at a restaurant.
The process used for decision and regression trees consists of using an algorithm that starts the top,
analyzes the data, and chooses the best variable on which to split the data. To create the split,
regression trees start by analyzing all input data and every possible split it could make. The split is
chosen to minimize the MSE of the prediction compared to the training data. This process is
repeated for every node until a decision is made. The decision is made when the mean square error
(MSE) of the node is less than the MSE for the entire data multiplied by the tolerance on quadratic
error per node, where the default tolerance is 10-6 (Matlab 2012). The Matlab statistics toolbox was
used for the calculations in this work.
In general these models are trained, i.e. the optimal weights are determined or DT structure
and assigned value found, by finding the weights or structure that makes the prediction algorithm
most closely match the training set. Several measures of can be used including absolute error,
squared error and mean square error amongst others. Unlike linear regression where explicit
formulas for the weights can be determined, determining the weights or structure for even a mildly
complex ANN or DT is often a complicated nonlinear optimization problem. There are a variety of
12 | P a g emethods to solve this problem, including steepest descent, conjugate gradient, quasi-Newton
methods and our choice the Levenberg-Marquardt algorithm, a method designed to utilize the best
parts of steepest descent and Newton based methods. A more expansive treatment of nonlinear
optimization can be found in Kelley (1999). As the relationship between the statistics and the point
differential is potentially highly nonlinear we need a reasonably sized ANN or DT to capture this
relationship. However the problem with complex networks is they can tend to overfit, i.e. fit the
data they are trained on very well at the expense of having little predictive capability. In addition as
the number of weights or size of tree to estimate increases the probability of finding a local
minimum greatly increases.
In order to overcome these difficulties people often use ensemble methods or a combination
of different learning algorithms. A common technique is to train many networks on random
portions of the total data set and then evaluate the individual algorithms by their ability to make
predictions on data they have never seen before. Using this one can avoid overfitting the training
data and having a single network with suboptimal weights. There are a large number of issues related
to how to assemble an ensemble of machines such as how many machines to use, whether the
individual machines have the same basic architecture, how do you eliminate less predictive machines
and how to combine the machine’s prediction once a final committee is chosen. We will not attempt
to discuss all these issues here but we will refer the reader to these surveys of methods for
combining networks into committees, as well as other related topics in Sharkey (1999) and Moreiraa
(2007).
Section 3. Results.
First, we use ANN’s and DTs to predict real interest rates. We used the following process:
13 | P a g e Calculate the mean RIR from all training data. Predict this as the RIR for all of our
test data. This is the simplest possible technique, in some ways similar to a
continuous version of flipping a coin.
Linear regression of our 13 ICRG variables in a year to predict the RIR in the same
year.
An ANN of our 13 ICRG variables in a year to predict the RIR in the same year.
A DT of our 13 ICRG variables in a year to predict the RIR in the same year.
We report the prediction accuracy of each method relative to a simple mean in this process in Table
1. We then use ensembles of ANN’s and DTs to predict real interest rates using the following
process:
Ensembles of 30 ANNs or DTs of our 13 ICRG variables in a year to predict the
RIR in the same year.
Ensembles of 30 ANNs or DTs using ICRG data from the current and previous 2
years and the RIR from the previous 2 years as well. This created the largest inputs
to our machine learning techniques for a total of 41 input variables.
In order to evaluate these techniques we used 80% of our data to build our models and a
final 20% to evaluate their predictive performance. We then looked at the mean absolute
error, the median absolute error and correlation between the predicted real IR and the actual
real IR. We then repeated this experiment 30 times and averaged each of these measures of
predictive ability.
Table 1 contains the predictive ability of each technique. Single model techniques,
including linear regression, ANNs and DTs performed poorly. In terms of mean and median
absolute error, they all performed worse than our simple benchmark prediction technique of
predicting the mean IR of the training data set. The ensemble techniques performed a good
14 | P a g edeal better, lowering all measures of error and increasing the correlation between the model
predictions and actual RIR data. The techniques that used ensemble techniques along with
the ICRG data and RIR from the past 2 years was the most predictive method, with the
ensemble of DTs being more accurate than the ANNs by a substantial amount.
15 | P a g eTable 2. : Summary of the various measures of predictive accuracy of all machine learning
techniques.
Mean over Mean Linear ANNs DTs
30 trials (simple Regression
predictor)
Mean Abs. 6.72 7.42 7.53 7.43
Error
Median 3.45 4.75 4.53 3.77
Abs Error
Correlation 0 0.29 0.28 0.43
Table 3. Summary of the various measures of predictive accuracy of all machine learning
techniques (ensembles).
Mean over Ensemble Ensemble Ensemble Ensemble
30 trials of ANNs of ANNs of DTs of DTs
(using past 2 (using past 2
years of years of
data) data)
Mean Abs 5.68 5.43 5.28 4.81
Error
Median Abs 3.40 3.13 2.85 2.45
Error
Correlation 0.60 0.60 0.62 0.66
The ensemble of DTs, which used both ICRG data and RIR rates from the previous
2 years, was the most predictive model. We therefore attempted to analyze the factors this
technique considered most important in making this prediction. As accurate DTs are
generally quite large and we are using ensembles of these we will not attempt to show a
picture of one of our DTs. Rather we will use a common technique in determining which
variables are most important in the prediction of a DT. To analyze the importance of each
16 | P a g eof our inputs for the DTs we permute the input variable with each other input variable and
average the increase in mean square error averaged over all DTs in the ensemble. Thus, if a
branch in a decision tree that was originally based on an important predictor is switched with
a decision based on an unimportant predictor we would expect the accuracy to decrease and
the mean square error to increase (Matlab 2012).
17 | P a g e'Money Growth-2'''
'Corruption'''
'Government Stability'''
'Corruption-2'''
'Government Stability-2'''
'Religion in Politics'''
'Ethnic Tensions-1'''
'Law and Order-1'''
'Religion in Politics-1'''
'Government Stability-1'''
'Law and Order'''
'Corruption-1'''
'Ethnic Tensions'''
'Religion in Politics-2'''
'Bureaucracy Quality-1'''
'Bureaucracy Quality'''
'Money Growth-1'''
'Law and Order-2'''
'Investment Profile-2'''
'Ethnic Tensions-2'''
'Democratic Accountability-1'''
Series1
'Internal Conflict-1'''
'Bureaucracy Quality-2'''
'Military in Politics-1'''
'Internal Conflict-2'''
'Military in Politics-2'''
'Democratic Accountability'''
'Socioeconomic Conditions'''
'External Conflict'''
'Military in Politics'''
'Internal Conflict'''
'Socioeconomic Conditions-2'''
'Socioeconomic Conditions-1'''
'Investment Profile'''
'Investment Profile-1'''
'External Conflict-2'''
'Democratic Accountability-2'''
'Money Growth'''
'External Conflict-1'''
'Real IR -2'''
'Real IR -1'''
0 0.5 1 1.5
Figure 2. Variable importance of all input variables in our ensemble of DTs ranked from
least important to most important.
18 | P a g eWe tabulate the relative importance of each variable, with two lags (the numbers in the
variable names refer to the lag length) in Figure 2. Notice that last year’s real interest rates are the
most important variable for predicting real interest rate. In a statistical sense this suggests a great
deal of autocorrelation. Normally this is a weakness of a statistical study and requires correction to
get statistically efficient and consistent results. But our atheoretical predictive technique turns this
weakness into predictive strength. Having said that, our ability to rank the other, institutional,
variables in order of importance as far as drivers of the real interest rate is concerned clearly suggests
that variability in institutions can be captured by real interest rates. Thus, while clearly money growth
matters for real interest rates, so does democratic accountability (with a two-year lag) and external
conflict (with a one year lag). First, of all the “institutional” variables included in our data, external
conflict is most important in predicting real interest rates. Conflict has a way of making life uncertain
and people impatient. On the other hand, democratic accountability matters as well if at a somewhat
slower rate since democratic accountability from two years ago is the next best variable for
predicting interest rates. An entrenched political leadership who can therefore be fearlessly
capricious presumably increases uncertainty and reduces patience; connecting democracy to interest
rates. At any rate, the reader will note here that we can establish a hierarchy of institutional variables
that help predict real interest rates. This directly fits with the idea that institutions that reduce free
riding and predation and move society towards a productive Pareto improving equilibrium are linked
with patience.
Section 4. Conclusion.
There is a significant theoretical literature that suggests that patience can resolve social free rider
problems that can lead to the evolution of Pareto improving cooperative equilibria. Another skein of
inquiry has looked at the effect of social and political institutions on economic growth. Taken
19 | P a g etogether the two strands of literature suggests that the sorts of institutions that help generate the
secular growth trend seen since the industrial revolution and societal patience levels – as expressed
in real interest rates – ought to be related. Indeed, we find that indices of institutional quality do
seem to predict real interest rates with a level of accuracy greater than a simple average. Of course, a
significant amount of the variation in real interest rates remains unexplained by the institutional
indices chosen by us in a limited data set of countries. However, we suggest that our study provides
a baseline for future work in this area. Adding more data would be extremely useful. Moreover, it
would be useful to look at the role of individual institutions in their effect on real interest rates. Last,
if institutions are good predictors of real interest rates then we suggest that real interest rates can
serve as a proxy for institutions. As a practical matter a business, for example, deciding to invest in a
country and trying to figure out the degree of institutional quality might compare the real interest
rate there relative to some OECD benchmark to help get some information in a less costly way than
that collected by the ICRG.
At the same time, we suggest that modern data mining techniques abstract away from many
of the endogeneity and data structure problems faced by a purely statistical analysis of the
relationship between interest rates and institutions. More generally, our methodological approach
might improve the problem of predictability in the social sciences.
Works Cited
Acemoglu, D. & James, A. R., 2012. Why Nations Fail: The Origins of Power, Prosperity, and Poverty. 1
ed. New York: Crown Publishing Group.
Basuchoudhary, A., Allen, S. & Siemers, T., 2010. Civilization and the evolution of short sighted
agents. Virgina Economic Journal, pp. 11-29.
Basuchoudhary, A., Mazumder, R. & Simoyan, V., 2012. The Evolution of Cooperation: How Patience
Matters. SSRN.
Basuchoudhary, A. & Razzolini, L., 2013. The Evolution of Revolution: Is Splintering Inevitable?.
Available at SSRN: http://ssrn.com/abstract=1731654 or http://dx.doi.org/10.2139/ssrn.1731654.
Buchanan, J. M., 1975. The Limits of Liberty: Between Anarchy and Leviathan. Chicago: University of
Chicago Press..
20 | P a g eClark, G., 2007. A Farewell to Alms: A Brief Economics History of the World (Princeton Economics
History of the Western World). Princeton: Princeton University Press.
Clark, G., n.d. The Interest Rate in the Very Long Run: Institutions, Preferences and Modern Growth.
[Internet] London: Centre for Economic Policy Research..
Dal-Bo, P. & Frechette, G. R., 2011. The Evolution of Cooperation in Infinitely Repeated Games:
Experimental Evidence. American Economic Review, Volume 101, pp. 411-429.
Diamond, J., 2005. Collapse: How Societies Choose to Fail or Succeed. New York: Viking Press.
Einav, L. & Levin, J., 2013. The Data Revolution and Economic Analysis. [Online]
Available at: http://www.nber.org/papers/w19035
[Accessed 13 February 2014].
Fudenberg, D. & Maskin, E., 1990. Evolution and Cooperation in Noisy Repeated. American Economic
Review, Volume 80, pp. 274-279.
Mueller, C. D., 1989. Public Choice II. Cambridge, U.K.: Cambridge University Press..
Mueller, D. C., 1989. Public Choice II. Cambridge: Cambridge University Press.
Nowak, M. & Highfield, R., 2011. SuperCooperators: Altruism, evolution, and why we need each other
to succeed. New York: Free Press (A Division of Simon and Schuster)..
Romer, D., 2001. Advanced Macroeconomics. 1 ed. New York: McGraw Hill.
Wright, R., 2001. Non Zero: The logic of Human Destiny. New York: Vintage Books.
21 | P a g eAppendix
Table 1. List of Countries.
Australia
Bahamas, The
Bangladesh
Bolivia
Botswana
Canada
Chile
China
Costa Rica
Gambia, The
Guatemala
Guyana
Honduras
Iceland
India
Israel
Italy
Japan
Kenya
Malawi
Malta
Nigeria
Oman
Papua New
Guinea
Philippines
Sierra Leone
Singapore
South Africa
South Korea
Switzerland
Syrian Arab
Republic
Thailand
Trinidad and
Tobago
United Kingdom
United States
Uruguay
Venezuela, RB
22 | P a g eYou can also read
