Chapter 1 Role of Probability in Engineering

1.1. Introduction
Quantitative methods of modeling, analysis, and evaluation are the tools of modern
engineering. Some of these methods have become quite elaborate and include sophisticated
mathematical modeling and analysis, computer simulation, and optimization techniques.
However, irrespective of the level of sophistication in the models, including experimental
laboratory models, they are predicated on idealized assumptions or conditions; hence,
information derived from these quantitative models may or may not reflect reality closely.
In the development of engineering designs, decisions are often required irrespective of the
state of completeness and quality of information, and thus must be formulated under
conditions of uncertainty, in the sense that the consequence of a given decision cannot be
determined with complete confidence. Aside from the fact that information must often be
inferred from similar (or even different) circumstances or derived through modeling, and thus
may be in various degrees of imperfection, many problems in engineering involve natural
processes and phenomena that are inherently random; the states of such phenomena are
naturally indeterminate and thus cannot be described with definiteness. For these reasons,
decisions required in the process of engineering planning and design invariably must be made,
and are made, under conditions of uncertainty.
The effects of such uncertainty on design and planning are important to be sure; however, the
quantification of such uncertainty, and evaluation of its effects on the performance and design
of an engineering system, properly should include concepts and methods of probability.
Furthermore, under conditions of uncertainty the design and planning of engineering systems
involve risks, and the formulation of related decisions requires risk-benefit trade-offs, all of
which are properly within the province of applied probability.

1.2 What Is Probability?

Probability is the branch of science concerned with the study of mathematical techniques for
making quantitative inferences about uncertainty. The key words in this definition are
quantitative and inferences. Indeed, as we will soon see, probability provides a mechanism for
making quantitative statements about uncertainty and, more important, allows one to draw
quantitative conclusions from such statements using the rules of logic.

Most historians consider the work of Fermat (1601-1665) and Pascal (1623-1662) on games
of chance to be the first significant contribution to the study of probability; however, many of
Fermat’s and Pascal’s ideas can be traced to earlier works of Cardan, Kepler, and Galileo.
There is also some evidence that the Romans, many centuries before, used mortality tables1 to
predict human lifespans. Since Fermat and Pascal’s timed, nearly every great mathematician
has some contribution to probability. Among the more famous contributors are the Bernoulli,
Laplace, DeMoivre, Poisson, DeMorgan, Venn, Bayes, Markov, and Kolmogorov.
While many scholars have studied probability purely for its intellectual and philosophical
appear, a good deal of the motivation for the subject has come, and continues to come, from
practical problems outside of mathematics. Indeed, the development of probability since
Fermat’s time has been heavily influence by the investigations in gaming, demography,
insurance, genetics, and quantum physics, to name just a few. Moreover, the subject itself has
had profound implications on everything from economics to engineered, it could be argued,
has played significant role in the history of the world over the last 200 years. To give a simple
example, consider marine insurance, whose issuance can be justified by the well-known law
of averages: the availability of marine insurance enabled commercial shipping to develop on a
large scale (because it freed maritime shippers from the worry of financial ruin due to a
catastrophe at sea), which in turn contributed to the economic and political ascendancy of
Britain in the 29tcentury and to international commerce as we know it.

Today, probability seed in a wide range of fields including engineering, finance, medicine,
meteorology, and management. We will encounter numerous applications of probability of
these and other fields throughout this book.

1.3 How Is Uncertainty Quantified?

If we agree that probability, from a scientific perspective, is the study of mathematical

techniques for making quantitative inferences about uncertainty, then for the subject to have
any meaningful content, we must have some precise way of quantifying uncertainty and
making inferences about that quantification. That there is considerable controversy over how
to precisely formulate such a quantification of uncertainty is an understatement, to say the
least. Indeed, some philosophers have gone so far as to argue that the very notion of
uncertainty cannot be precisely quantified since to do so would, in effect, make uncertainty
certain.

One approach to uncertainty is to use the concept of relative frequency. To describe this
concept, consider an experiment with several possible outcomes which can be replaced a large
number of times. The relative frequency of a particularly outcome of such an experiment in a
sequence of repetitions of the experiment is the fraction of the total number of repetitions of
the experiment that result in the desired outcome, for example, in the sequence of coin losses
resulting in H, T, T, T, H (where H0 signifies heads and T signifies tails), the relative
frequency of heads is 2/5, whereas in the sequence of tosses resulting in T, H, T, H, H, the
relative frequency of heads is 3/5. Experience suggests that as the number of repetitions of the
experiment increases, the relative frequencies associated with a particular outcome coverage
to a common value. For example, the relative frequency of heads approaches ½ as the number
of coin tosses increases, provided that the coin is not biased. This common value to which the
relative frequencies converge is called the probability of the desired outcome.

This approach to quantifying uncertainty, while intuitively appealing has some major
drawbacks, the most serious of which is the reliance on the ambiguous notion of a limiting
relative frequency. The early probabilists overcame logical difficulties by restricting their
attention to experiments in which the number of outcomes is finite and by assuming that all
outcomes of such experiments are equally likely, (i.e., have the same probability). While the

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 1

assumption of equal likelihood of outcomes is admittedly idealized, in the context of the
games of chance with which the early probabilists were concerned, it is not unrealistic
provided that one correctly identifies the outcomes of the experiment s captured and no
information is suppressed.

To illustrate the difference between a correct and an incorrect application of the principle of
equal likelihood, consider the experiment in which two unbiased coins are tossed. If one
identifies the possible outcomes as being head-head, head-tail, tail-head and tail-tail, then one
correctly assigns a probability of ¼ to each of these outcomes and one correctly deduces that
the probability of getting exactly one head is ½. However, if one fails to distinguish between
the coins and identifies the possible outcomes as two heads, one head-one tail, and two tails,
then one incorrectly assigns a probability of 1/3 to each of the outcomes head-head, tail-tail,
and one incorrectly deduces that the probability of getting exactly one head is 13.

The great achievement of the classical probabilities was to initiate a logical approach to the
study of uncertainty, which to a great extent is still with us today. By avoiding the difficulty
inherent in considering probabilities tot be limiting relative frequencies and instead assuming
that all experimental outcomes are equally likely, they were able to focus their energies on
developing a logical system for deducing the probabilities of particular groups of observations
that were often too difficult to determine accurately by successive repetition of an experiment.
This system of logical deduction also enabled them to avoid being misled by potentially faulty
intuition.

Table 1.1 Results from 100 Repetitions of a “Ten Coin Toss:

______________________________________________
Number of Heads Frequency
______________________________________________
0 0
1 1
2 3
3 12
4 21
5 24
6 22
7 13
8 4
9 0
10 0
_____
100
________________________________________

To give a simple illustration of a situation in which the deductive approach succeeds here
intuition might fail, consider the probability of getting exactly five heads in ten tosses of a
coin. One might think that if the coin as likely to heads as it is to land tails, then this
probability should be 1/2 since, according to the relative frequency interpretation of

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 2

probability, an unbiased coin that is tossed a large number of times will land heads
approximately half the time. However, if you thought this, you would be wrong! In fact, under
the assumption that the probability of getting heads in a single toss is 50%, one can show
deductively that the probability of getting exactly five heads in ten tosses of the coin is 63/256
= 0.24609375. Interestingly enough, the value 63/256 is in accord with the relative frequency
interpretation of probability, as one can confirm tby repeatedly tossing a fair coin ten times
and computing the corresponding relative frequencies. the results of 100 such repetitions are
given in Table 1.1.

The deductive approach to probability taken by the classical probabilities is an example of the
axiomatic approach. The axiomatic approach in mathematics is a deductive technique in
which the topic of interest is described by a collection of axioms in the language of sets, and
all inferences about the topic are made using only these assumptions and the rules of set
theory and formal logic. Mathematicians struggled for many years to find an axiomatic
formulation for probability that would encompass all types of experiments, not just the one
considered by the classical probabilists. Finally, in the 1903s, the Russian mathematician A.N.
Kolmogorov gave an axiomatic description for the theory of probability that permitted
virtually every experiment to be considered.

In most applications, an intuitive understanding of probability based on relative frequencies is

generally sufficient. However, it is nice to know that the subject rests on a firm foundation
and that the conclusions we reach have some basis in logic!

Before moving on to the next section, it is instructive to make one more remark about the
meaning of probability statements. While on often makes probability statements about
individual outcomes of an experiment, it is important to keep in mind that such statements are
assertions about groups of observations. For example, when a doctor tells a patient who must
undergo a lifesaving operation that the probability of survival is, for instance, 80%, the doctor
is really making a statement about a group of similarly situated individuals because such
probability statements have little or no meaning for individuals in isolation. After all, any
given individual either survives the operation or does not survive.

1.4 Uncertainty in Real-World Information

Many phenomena or processes of concern to engineers contain randomness; that is, the actual
outcomes are (to some degree) unpredictable. Such phenomena are characterized by
experimental observations that are invariably different from one experiment to another (even
if performed under apparently identical conditions). In other words, there is usually a range of
measured or observed values; moreover, within this range certain values may occur more
frequently than others. The characteristics of such experimental data can be portrayed
graphically in the form of a histogram or frequency diagram. The natural state of most
engineering information contains significant variability. The histogram, therefore, is a graphic
empirical description of the variability of experimental information. For a specific set of
experimental data, the corresponding histogram may be constructed.
Table 1.1 Year wise Rainfall Intensity

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 3

 Year Rainfall intensity (in.)
1918 43.30
1919 53.02
1920 63.52
1921 45.93
1922 48.26
1923 50.51
1924 49.57
1925 43.93
1926 46.77
1927 59.12
1929 47.38
1930 40.78
1931 45.05
1932 50.37
1933 54.91
1934 51.28
1935 39.91
1936 53.29
1937 67.59
1938 58.71
1939 42.96
1940 55.77
1941 41.31
1942 58.83
1943 48.21
1944 44.67
1945 67.72
1946 43.11

Plotting the number of observations in a given rainfall interval, we obtain the histogram of the
rainfall intensity in the region on interest, as shown in Fig.1.1a, whereas, in terms of the
fraction of total observations the same histogram would be as shown in Fig.1.1b.

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 4

 Fig. 1.1 Histograms of representative rainfall intensity

Table 1.2 Fraction of observations vs. Interval

Interval Number of observations Fraction of total observations
38-42 3 0.1034
42-46 7 0.2415
46-50 5 0.1724
50-54 5 0.1724
54-58 3 0.1034
58-62 3 0.1034
62-66 1 0.0345
66-70 2 0.0690
Total=29 1.000

For the purpose of comparing an empirical frequency distribution (as, for example, portrayed
in a histogram) with a theoretical probability density function, the corresponding frequency
diagram is required.

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 5

The histogram, or frequency diagram, gives a graphic picture of the relative frequencies of the
various observations or measurements. For most engineering purposes, certain aggregate
quantities from the set of observations are more useful than the complete histogram; these
include, in particular, the mean-value (or average) and a measure of dispersion. Such
quantities may be evaluated from a given histogram; statistically, however, these are usually
obtained in terms of the sample mean and sample standard deviation.

Clearly, if recorded data of a variable exhibit scatter or dispersion, such as those illustrated in
the value of the variable cannot be predicted with certainty. Such a variable is known as a
random variable, and its value (or range of values) can be predicted only with an associated
probability.

When two (or more) random variables are involved, the characteristics of one variable may
depend on the value of the other variable (or variables). Pairs of observed data for the two
variables, when plotted on a two-dimensional space are characterized by scatter or dispersion
in the data points, called scattergrams. In view of such scatter, the value of one variable,
given that of the other, cannot be predicted with certainty. The degree of predictability will
depend on the degree of mutual dependency or correlation between the variables, as measured
(in the linear case) by the statistical correlation. Methods for evaluating such correlations are
also embodied in statistical analysis.

It should be strongly emphasized that the application of stochastic methods is not limited to
the description of experimental data, or to the evaluation of the associated statistics (such as
the mean, standard deviation, and correlation). Indeed, the much more significant role of
probability concepts is in the utilization of this information in the formulation of proper base
for decision making and design. In other words, when we are dealing with information which
requires probabilistic description, the proper utilization of this information in engineering
design and planning will necessarily require concepts and methods of probability. For
example, if a design equation involves random variables, the quantitative analysis of their
effects on, and the formulation of, the design will necessarily involve probabilistic concepts.

1.5 Uncertainty associated with imperfect modeling and estimation

Engineering uncertainty, however, is not limited purely to the variability observed in the basic
variables. First, the estimated values of a given variable (such as the mean) based on
observational data will not be error free (especially when data are limited). In fact, in some
cases, such estimates may not be much better than "educated guessers," based largely on the
engineer's judgement. Second, the mathematical or simulation models (for example, formulas,
equations, algorithms, computer simulation programs), and even laboratory models, that are
often used in engineering analysis and to develop designs are idealized representations of
reality; in various degrees, such models are imperfect representations of the real world.

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 6

Consequently, predictions and/or calculations made on the basis of these models may be
inaccurate (to some unknown degree) and thus also contain uncertainty. In certain cases, the
uncertainties associated with such prediction or model errors may be much more significant
than those associated with the inherent variabilities.

All uncertainties, whether they are associated with inherent variability or with prediction
error, may be assessed in statistical terms, and the evaluation of their significance on
engineering design accomplished using concepts and methods that are embodied in the theory
of probability.

1.6 Design and Decision Making Under Uncertainty

If information is of the type illustrated in histograms in which no single observation is
representative, and evolutions and predictions must be based on imperfect models, how
should designs be formulated or decisions affecting a design be resolved? Presumably we may
assume consistently worst conditions (or example, specify the highest possible flood, smallest
observed fatigue life of materials, and so on) and develop conservative designs on this basis.
From the standpoint of system, performance and safety, this approach may be suitable;
however, the resulting design could be too costly as a consequence of "compounded
conservatism." On the other hand, an inexpensive design may not ensure the desired level of
performance or safety. Therefore, decisions based on trader-off between cost and benefit
(including tangible and intangible factors) are necessary. The most desirable solution is one
that is optimal, in the sense of minimum cost and or maximum benefits. If the available
information and evaluative models contain uncertainties, the required trade-off analysis
should include the effects of such uncertainties on a given decision.

Such situations are common to many problems of engineering design and planning; in this
section we describe several examples illustrating some of these problems. The examples are
idealized to simplify the discussions; nevertheless, they serve to illustrate the essence of the
decision-making aspects of engineering under conditions of uncertainty.

1.6.1 Planning and design of airport pavement

Consider, as the first example, the design of an airport pavement. Among the many factors
that have bearing on the design, the thickness of the pavement system (consisting of several
layers of subgrade base material and the finished pavement) is one of the principal decision
variables. In general, the usable life of the pavement will depend on the thickness of the
system; the thicker the pavement system is, the longer its useful life will be. Of course, for the
same material and workmanship quality, the cost will also increase with the thickness. on the
other hand, a thin system will cost less initially, but the subsequent maintenance and
replacement costs will be higher. Therefore, the thickness of the pavement system may be
determined on the basis of a trade-off between high initial cost will low maintenance, versus

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 7

low initial cost but high replacement and maintenance costs. For the purpose of such trade-off
analysis, the relation between the life of a pavement system and its thickness is required.
However, the pavement life is also a function of other variables, including drainage and
moisture content, temperature ranges, density and degree of compaction of the subgrade.
Since these factors are random (as seen in the histograms above), the life of the pavement
cannot be predicted with certainty. Hence, the total cost (including initial and maintenance
costs) associated with a given pavement thickness cannot be estimated with complete
confidence; any meaningful trade-off analysis, therefore would properly require concepts of
probability.

1.6.2 Design of structures and machines

Consider, for example, the design of an offshore drilling tower, which is subject to occasional
hurricane forces. In such a case, we recognize that aside from the fact that the maximum wind
effect during a hurricane is random, the occurrence of hurricanes in a given coastal region is
also unpredictable. Hence, in determining the safety level for the design of the tower, the
probability of occurrence of strong hurricanes within the specified useful life of the structure
must be considered, addition to the survival probability of the structure during a hurricane.
The higher the hurricane force gets, the less frequent will be its occurrence therefore, if a very
strong hurricane is specified for the design, there may be almost no chance of it occurring
during the useful life of the drilling tower. Consequently, what level of hurricane force should
be used in the design, and what level of protection would be adequate during a hurricane are
decisions that clearly require trade-offs between cost and level of protection in terms of risk or
failure probability within the life time of the structure.

In structural or machine components that are subject to repeated cyclic loads, the fatigue life
(that is, the number of load cycles until fatigue failure or fracture) of the component is also
random, even at constant amplitude stress cycle, as illustrated in several equivalent
histograms. For this reason, the useful life of the component is, to some degree, unpredictable.
A design will depend on the required life and desired level of reliability; for a given design,
the shorter is the required service life, the higher will be its reliability against possible
breakdowns within the specified service life. Fatigue life is also a function of the applied
stress level; generally, the higher the stress, the shorter the fatigue life. Consequently, if a
desired life is specified the components could be designed to be massive so that the maximum
stresses will be low and thus assure long life. This approach will, of course be expensive in
terms of material. In contrast, if the parts are under designed, high stresses may be induced,
resulting in short life and thus requiring frequent replacements.
The optimal life may be determined on the basis of minimizing the total expected cost, which
would include the init8ial cost, the expected cost of replacement (a function of the reliability
or probability of no failure), and the expected cost associated with the loss of revenue

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 8

incurred during repair (also a function of reliability). Having decided on the desired design
life, the components may then be proportioned accordingly.

1.7 Other examples of application of probability in engineering and the sciences

Probability can also be applied to specific areas of engineering such as reliability, quality
control, and the analysis of queues, insurance, financial market behavior etc.
Reliability is the branch of engineering concerned with the lifetimes of electrical and
mechanical systems. Probability can be used to analyze the uncertain lifetimes of such
systems and to assist the engineer in designing systems that are both reliable and cost
effective.
Quality control is the branch of engineering concerned with the maintenance of quality in
manufacturing processes. The goal of quality control is to minimize the number of defects
produced by a manufacturing process without incurring unreasonable costs and to identify
defects before they leave the production line. Quality control usually involves testing items in
the production line to ensure that appropriate standards are met. Since it is not cost effective,
and in some cases not even practical, to test every item on a production line, a sample of items
is usually tested, and inferences about the number of defectives produced in total are made on
the basis of test results. Probability can be used to analyze the uncertainty in these inferences
and to determine the size of the sample to be tested.
Queuing is the branch of engineering concerned with the analysis and design of systems
involving multiple servers and multiple clients, in which clients may be required to wait for
service. The simplest example of a queuing system is the checkout line at the grocery store or
the service line at a bank or post office. Examples of queues can also be found in multiuser
distributed computing systems (so called client-server systems). Probability can be used to
analyze the uncertain waiting and service time in such queues and to assist the engineer or
computer scientist in designing a system that meets the demands imposed on it in the most
cost-effective way.

1.8 Actuarial Science

Outside of the insurance industry, relatively few people have ever heard of actuarial science.
Actuarial science is the subject whose primary focus is analyzing the financial consequences
of future uncertain events. In particular, it is concerned with analyzing the adverse financial
consequences of large, unpredictable losses and with designing mechanisms to cushion the
harmful financial effects of such losses. Among these are the loss or destruction of property
due to fire, theft, or natural disaster, the loss of employment due to an economic contraction
or the obsolescence of one’s skills; and the loss of health due to accident, sickness, or injury.
Insurance systems have evolved to cushion the effects of such large, unpredictable losses.
Insurance is based on the premise that individuals faced with large and unpredictable losses
can reduce the financial effects of such losses by forming a group and sharing the losses
incurred by the group as a whole.

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 9

Consider, for example, a group of homeowners who individually risk having their homes
destroyed by a tornado. Clearly, the risk of loss faced by each individual in the group is
substantial even if tornadoes are a rare occurrence because the size of a loss, when it occurs,
will generally be very large. However, the financial consequences of such losses can be
reduced through sharing; indeed, if the members of the group agree to pay an equal portion of
the total loss incurred by the group as a whole, then the amount that each individual will be
required to pay becomes more certain and the chance that any one person will be financially
responsible for a large loss becomes very small.

This important principle of loss sharing, known as the insurance principle, forms the
foundation of actuarial science. It can be justified mathematically using the law of large
numbers from probability theory. For the insurance principle to be valid, essentially four
conditions should hold (or very nearly hold):
1. The losses should unpredictable.
2. The risks should be independent in the sense that a loss incurred by one member of
the group makes additional losses by other membership of the group no more or less
likely.
3. The risks should be homogeneous in the sense that a loss incurred by one member the
group is not expected to be any different in size or likelihood from losses incurred by
other members of the group.
4. The group should be sufficiently large so that the portion of the total loss that each
individual is required to pay becomes relatively certain.

Let’s consider each of these conditions in turn.

Losses Should Be Unpredictable. If the time and sized of a future loss are known in advance,
then apart from finding a charitable benefactor, there is nothing one can do to moderate the
financial effects of the loss. For it is unlikely that anyone would be willing to share the
expense of such a loss unless they themselves were certain to have a future expense of equal
or greater size, in which case there would be no benefit to entering into a loss-sharing
arrangement. Hence, when losses are not unpredictable, insurance is ineffective.

Risks Should be Independent. If the losses incurred by individuals in the group tend to occur
in concert with one another (i.e., if they are dependent), then the portion of the group’s total
loss paid by each individual will not be appreciably smaller than the large individual loss that
would have to be paid by the individual member in the absence of a loss-sharing arrangement.
For instance, if the homeowners in the previous example all live on the same street, then it is
quite likely that when a tornado strikes the neighborhood, it will damage many of the houses
in the group and the resulting repair cost to each member will still be potentially ruinous even

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 10

though the total repair cost is spread over the entire group. Hence, when risks are not
independent, insurance can be ineffective.

Risks Should Be Homogeneous. If the losses expected to be incurred by each member of the
group are not the same, then the members with the lower or the less likely losses will not
consider the equal distribution of loss expenses to be fair and will not agree to pay for the
losses of the other members of the group. For instance, if some of the homeowners in the
previous example had houses that were much more expensive to replace than the rest or were
located in an area where tornadoes occurred more frequently, then it is unlikely that the
homeowner with the more modest or more protected homes would be willing to share equally
in the cost of reconstruction in the event of a tornado. Hence, when risks are not
homogeneous, insurance arrangements can break down.

Group Should Be Sufficiently Large. If the number of individuals in the group is too small,
then the portion of the total loss that each individual in the group is required to pay will be
highly unpredictable and is likely to be prohibitively large. In this case, loss sharing will offer
the individual little improvements over bearing a loss alone and could in some cases result in
a worse situation because the chance that an individual will have to make a payment is
actually greater when the individual is part of a group. Hence, when the group is not
sufficiently large, insurance is not very effective.

Each of these four conditions has a precise mathematical formulation in the language of
probability. In practice, however, risks are not truly independent or homogeneous. Moreover,
there will always be situation where the condition of unpredictability is violated. However, as
long as the dependence among the risks is weak and the differences among the risks are not
too great to cause large inequities in the amounts individuals are required to pay, insurance is
generally feasible. In situations where heterogeneous (i.e., non-homogeneous) risks must be
combined to obtain a group that is sufficiently large, an equitable distribution of loss expenses
can generally be achieved by appropriately weighing the amount payable by each member of
the group.

It is worth pointing out that insurance arrangements such as the one just illustrated cannot
eliminate misfortune; indeed, homes are still going to be destroyed by tornadoes. However,
insurance arrangements can make the financial effects of such misfortune less devastating and
more certain. This is a common feature of insurance. Many people mistakenly believe that
insurance is akin to gambling. However, this is not so because gambling creates a risk where
there was none before, whereas insurance manages an existing risk that is unavoidable.

For practical reasons, people seeking insurance do not generally try to form loss-sharing
groups by themselves. Instead, they turn to financial intermediaries such as insurance

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 11

companies to do this for them. In particular, people seeking insurance protection enter into
contracts with insurance companies in which they transfer the unwanted risk to the insurance
company and agree to pay the insurance company an up-front premium for this service. By
entering into a large number of such contracts with different individuals, the insurance
company can, in turn, form a group for which the insurance principle holds, thereby ensuring
that its aggregate claim expenses are reasonably predictable, and make a profit in the process.
It is at this juncture that actuarial science enters the picture.

Actuarial science seeks to address the following three problems associated with any such
insurance arrangement:
1. Given the nature of the risk being assumed, what price (i.e., premium) should be insurance
company charge?
2. Given the nature of the overall risks being assumed, how much of the aggregate premium
income should the insurance company set aside in a reserve to meet contractual
obligations (i.e., pay insurance claims) as they arise?
3. Given the importance to society and the general economy of having sound financial
institutions able to meet all their obligations, how much capital should an insurance
company have above and beyond its reserves to absorb losses that are larger than
expected? Given the actual level of an insurance company’s capital, what is the
probability of the company remaining solvent?

These are generally referred to as the problems of pricing, reserving, and capital allocation
and insolvency.

Actuaries must also be concerned with two economic problems, adverse selection and moral
hazard, which arise from an insurance company’s inability to access perfect information
about people purchasing insurance. Economists refer to this more general problem as the
problem of imperfect information. Adverse selection in an insurance context arises from an
inability to distinguish completely the good risks from the bad. If an insurer is unable to
distinguish between risks and charges a uniform premium for all, then the good risks will find
the insurance protection expensive and let their policies lapse, whereas the bad risks will find
the insurance protection a bargain and purchase more. The result will be larger claim
payments than anticipated and inadequate premium income to cover the payments. Moral
hazard in an insurance context arises from the behavioral changes that insurance protection
induces after it is purchased: Once insured, people are more likely to act in reckless ways than
they otherwise would because they know that the insurance company will pay for any losses.
These two problems can have a significant impact on pricing, reserving, and solvency.
Actuaries generally try to minimize their impact by designing products with features such as
deductibles and coinsurance, which serve to align the interests of the policyholder with those

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 12

of the insurer and encourage policyholders through their choice of policy to reveal
information that enables insurers to better determine their risk class.
1.9 What Is Financial Engineering?
Financial engineering is a relatively new discipline that is focused on the analysis of risk in
financial markets and the design of products and techniques to manage that risk. This rapidly
evolving field, which first came into prominence in the 1990s, relied heavily on the use of
sophisticated computers and computational techniques to exploit market anomalies and
manage financial risk.
Four factors have contributed to the development of financial engineering: technology,
deregulation, globalization, and the increased reliance on capital markets and market forces in
public policy.
• Rapid advances in computer and communications technology since the later 1970s
combined with the resulting fall in transaction costs have made practical the
implementation of sophisticated investment strategies such as program trading, portfolio
insurance, dynamic hedging, and portfolio replication.
• Deregulation prohibiting commercial banks, investment banks, and insurance companies
from entering each other’s businesses, has enabled the development of new financial
products combining savings, investment, and insurance features which until now were
only theoretically possible.
• Globalization of financial markets and of business has led to the development of global
financial institutions and of financial products (e.g., cross-currency options) for managing
the risks associated with doing business in different countries with different legal systems
and different currencies.
• Finally, the increased reliance on capital markets and market forces as illustrated through
the privatization of state-owned enterprises in former socialist countries and the
privatization of public pension schemes in Latin America has stimulated the demand for
new financial products (e.g., mutual funds with return guarantees) to manage risks
previously assumed and obscured by government, but now made explicit by the capital
market.
The fundamental principle underlying much of financial engineering is the principle of no
arbitrage. This principle asserts that two securities that provide the same future cash flow and
have the same level of risk must sell for the same price. Equivalently, the principle asserts that
a risk-free investment of zero can only have a return of zero. In other words, there is “no free
lunch”.
Arbitrage opportunities can arise from time to time. However, the systematic exploitation of
such opportunities by market players will cause market prices to adjust until a point is reached
where the arbitrage opportunity no longer exists or it is no longer profitable to take advantage
of the opportunity due to the size of transaction costs. Hence, the principle of no arbitrage
implicitly assumes a well-functioning market that is in equilibrium.

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 13

A related concept that arises frequently in financial engineering is the principal of
optimality. This principle asserts that investors, when forming investment portfolios, should
allocated their funds among the available securities in a way that optimizes the portfolio’s
risk-adjusted return.
Financial engineers combine these two principles with techniques from probability, statistics,
mathematical physics, and numerical computation to construct optimal portfolios and manage
financial risk.
One may notice the similarity between the work of financial engineers and the work of
actuaries. Indeed, it would appear that both types of professionals are concerned with the
financial consequences of future uncertainties, which is our definition of actuarial science.
The current differences between the disciplines lie in the types of future uncertainties
considered (e.g., risk of premature death vs. risk of a market crash) and in the principles used
(e.g., insurance principle vs. principle of no arbitrage). There is also a historical difference:
Actuarial science traditionally was concerned with uncertainty in the liabilities of the
business, whereas financial engineering developed in response to increasing uncertainty in the
assets of the business, which had previously been ignored. With the trend by businesses to
consider contingencies in their assets and liabilities together, it appears likely that the
disciplines of actuarial science and financial engineering will move closer together, and it is
possible that they will eventually merge.

1.6 Interpretations of Probability

So far probabilities have been considered to be relative frequencies and it has been tacitly
assumed that they are fixed even though our estimate of them may change from one set of
experimental data to another. This perspective, in which probability is considered to be a
constant long-run relative frequency, is known as the frequentist or objectivist interpretation
of probability. However, there is another perspective, known as the Bayesian or subjectivist
interpretation, in which probabilities are considered to be measures of personal belief. With
the Bayesian perceptive, the probability assigned to a particular event can be different for two
different people even if it is based on the same set of experimental data and can change over
time to reflect new information and evolving opinion.
Both interpretations of probability have coexisted since 17th century. Indeed, the classical
probabilists’ assumption of equal likelihood of outcomes could be considered an expression
of their belief in the fairness of the games that they were studying. However, from time to
time, one or the other of these perspectives has dominated the thinking of the scholars of the
day. For example, during the first half of the 20th century, the frequentist perspective so
dominated statistical thinking that anyone suggesting a Bayesian approach to a problem risked
becoming an outcast in the scholarly community. Today, both perspectives are generally
regarded as valuable. However, this does not mean that the Bayesian interpretation is not
subject to debate, sometimes heated!

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 14

To fully appreciate the difference between the frequentist and the Bayesian perspective, and
to help you form an opinion about which viewpoint is “more correct,” it is instructive to
consider a concrete example. Hence, suppose that you are given a coin with two
distinguishable sides to be used in some game of chance, and consider how you would
interpret the probability of getting heads in each of the two paradigms, both before and after
being given historical data on the coin.
Frequentist Perspective. Before being told anything about the coin’s history, the frequentist
observes that there are two possible outcomes, assuming that the coin always lands on one
side or the other, but reasons that nothing can be said about the probability of heads since
there are no available data on which to make a statement about the long-run relative
frequency. Hence, the frequentist concludes only that the probability lies between zero and
one, nothing else. If it is then revealed that in 1000 tosses of the coin 550 heads were
observed, the frequentist concludes that the estimate for the probability of heads is 55%,
acknowledging that the true probability remains unknown. If it is further revealed that in a
different 1000 tosses of the coin 510 heads were observed, the frequentist concludes (nothing
that a total of 1060 heads have been observed in 2000 tosses) that while the unknown
probability of heads is the same as before, the estimate for this probability should now be
53%. For the frequentist the estimate of the probability of heads can change, but the
probability of heads itself remains constant. Hence, according to a frequentist, the probability
of heads is an inherent constant that does not change with the arrival of new data.
Bayesian Perspective. Before being told anything about the coin’s history, the Bayesian
observes that the coin is symmetrically constructed except for the different markings on the
two sides of the coin and reasons, by the laws of physics, that there should not be a tendency
for the coin to land on one side more than the other. Hence, in the absence of historical data,
the Bayesian is led to believe that the probability of heads is 50%. If it is then revealed that in
1000 tosses of the coin 550 heads were observed, the Bayesian reason that the probability of
heads is closer to 55% and acknowledges that the original belief about this probability was
incorrect. If it is further revealed that in a different 1000 tosses of the coin 510 heads were
observed, the Bayesian then reasons that the probability is closer to 53% and acknowledges
again that earlier beliefs about this probability were incorrect. For the Bayesian, both the
estimate of the probability and the belief about what that probability was in the first place can
change over time. Hence, according to a Bayesian, the probability of heads is not an inherent
constant; rather, it is a number that can change over time to reflect that experimenter’s
opinion.
In a more sophisticated Bayesian analysis, one can consider the probability p of getting heads
to be a “random” quantity itself, and one can assign “probabilities” to each of the values
between zero and one in a way which reflects one’s belief that the unknown probability p is a
particular value; over time, this assignment of probabilities to the possible values of p changes
with the arrival of new information to reflect changes in one’s belief about p. For example,

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 15

prior to seeing any data, one may believe that p is within 1% of .50 “with high probability”
(e.g., with 80% probability); however, after learning that 550 heads have been observed in
1000 tosses, one may alter this opinion and instead believe that there is only a 25% chance
that p is within 1% of .50. The key point to remember is that the notion of probability as it
applies to p is a measure of belief that varies from person to person.
Classical statisticians argue that Bayesian methods suffer from a lack of objectivity because
different individuals are free to assign different probabilities to the same event according to
their personal opinions. Bayesians counter that the classical methods, based on a frequentist
interpretation of probability, have built-in subjectivity (e.g., through the design of an
experimental sampling procedure) and that the advantage of the Bayesian approach is that the
subjectivity is made explicit.

CL 425 Ch 1: Stochastic Methods in Engineering Planning and Design 16