Risk theory Decision-making
under uncertainty



This glossary explains some terms that are used on the pages presented here. This is not a complete dictionary, but rather a collection of brief descriptions.


A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

A B
  • Bernoulli distribution. Discrete probability distribution; corresponding random variable takes only two values, usually 0 and 1. More.
C
  • Cauchy distribution. Continuous probability distribution with heavy tails. This distribution is used to model situations when large deviations from "mean" value are possible. More.
  • Certainty equivalent of a random variable X. In utility theory this means the quantity possessing the same utility as X. Given utility function U, certainty equivalent is calculated as
    d(X) = U-1(EU(X)),
    where U-1 is inverse function for U. Certainty equivalent may be used as a risk price in risk exchange transactions.
  • Concave function. A real function f is called concave (strictly concave), if a function (-f) is convex (strictly convex).
  • Convex function. A real function f is called convex, if for any x, y in its domain and any number a from (0,1) the following inequality is true:

    f (ax + (1 - a)y) <= a f(x) + (1 - a) f(y).

    If inequality is strict, then f is called strictly convex. Convexity topics are discussed in detail in the book R. Rockafellar. Convex Analysis. Princeton univ. press, Princeton, 1970.

  • Copula. A multidimensional distribution function on a hypercube [0,1]n with uniform marginals. By Sclar theorem any multidimensional dstribution may be represented by superposition of a copula and marginal distributions, thus copula completely describes dependence of components.
  • Correlation of random variables X, Y is a measure of their linear dependence. Correlation is calculated as
    r = E[(X - EX)(Y - EY)] / (s(X) s(Y)),
    where s(X), s(Y) are standard deviations of X, Y. Correlation is especially useful under normal joint distribution of X, Y, while under other distributions may be misleading, and should be used with caution.
D
  • Distribution density. A charactiristic of a continuous distribution, that is equal to derivative of a cumulative distribution function. More.
  • Diversification. Distribution of resources, say, investment capital, among several possibilities or projects (investment tools). Used to decrease riskiness of a portfolio obtained. In second order theories diversification is measured in terms of expectation and variance of a portfolio. Modern theories used to measure riskiness by applying risk measure concept.
E
  • Efficient frontier. In a vector optimization problem: a set of points on criteria plane corresponding to efficient solutions to a problem. By changing an efficient solution one can improve value of some criterion only at the expense of worsening value of another criterion. In a Markowitz problem efficient frontier is represented by an upper branch of parabola that is a boundary for admissible solutions on risk-return plane.
  • Expectation of a random variable - its mean value. Expectation of number of points in tossing a dice is equal to 3 1/2. More.
  • Expected utility. A generalization of utility concept from certain to random objects. Expected utility is calculated as expectation of utility of random object of interest. Let U be an utility function, X be a random variable, describing, say, a return of investment instrument. Then expected utility of X is calculated as
    u(X) = EU(X).
  • Exponential distribution. Continuous distribution on positive reals. This distribution is often used in reliability theory to describe life time of technical elements, and sometimes is used in insurance models to describe human life time. More.
F
  • Frechet bounds. Any multidimensional distribution function
    FXY...Z(x,y,...,z)
    with marginals FX(x), FY(y), ..., FZ(z) satisfies inequality
    FXY...Z(x,y,...,z) <= min{FX(x), FY(y),..., FZ(z)},
    that is call upper Frechet bound. For two-dimensional distribution the lower Frechet bound is also valid:
    FXY(x,y) >= max{0,FX(x)+FY(y)-1}.
G
H
I
  • Independent random variables. Random variables X, Y are called independent if their distribution functions функций распределения satisfy
    FXY(X <= x; Y <= y) = FX(X <= x) FY(Y <= y).
    Otherwise the random variables are called dependent.
  • Independent events. Random events A, B are called independent if their probabilities satisfy
    P(AB) = P(A) P(B).
    Otherwise the events are called dependent.
J
K
L
  • Lognormal distribution. This distribution is often used to describe return of financial instruments, since it takes only positive values. More.
M
  • Median of a random variable is one of its mean characteristics. More.
  • Mode of a random variable is one of its mean characteristics, represents most probable value. More.
  • Monte Carlo method. A general method for solving problems, if analytic solution is difficult or does not exist. The essence of the method is representing a solution by probability distribution or a functional of probability distribution, and then solving the problem by sequencial sampling from the distribution. Detailed description of the method is given in the lecture.
N
  • Normal distribution. The most widely used probability distribution. Sometimes assumption of normality may be greatly misleading, so it should be used with caution. More.
O
  • Outcomes space of a probabilistic experiment: a set X={x,y,z,...} of all possible states that object of interest may fall into as a result of experiment.
P
  • Portfolio. A result of distribution of a resource (investment capital) among several possibilities (projects). For n projects desribed by a random vector of returns
    (X1,...,Xn),
    portfolio is defined by weights or portions
    y=(y1,...,yn)
    of investment capital being invested in each project. Portfolio return is a random variable
    X=y1X1+...+ynXn,
    so portfolio selection problems are usually stated as achieving the "best" (in some sence) portfolio distribution by choosing weights subject to rstriction
    y1+...+yn=1.
  • Probability of event. Degree of likelihood of event, a measure on some probability space. Takes numeric values between 0 and 1. In experiment with tossing a dice the probability of appearing of each side is equal to 1/6.
  • Probability distribution. In a simple case the set of all outcomes X={x,y,z,...} of probabilistic experiment is finite (consists of n elements). Probability distribution is a collection of n nonnegative numbers ( probabilities) px, py, pz,..., that sum to 1.
Q
R
  • Random variable. (Measurable) mapping from outcomes space of a fixed probability experiment into reals. Random variable serves a probabilistic model for experiments with uncertain number outcome. Example is number of points in tossing a dice. This random variable takes values 1, 2, 3, 4, 5, 6 with probability 1/6 each. The more frequently used parameters of random variable are expectation and variance.
  • Risk. 1. Probabilistic uncertainty about future events.
    2. Mathematical object - a probability distribution of an abstract random element, e.g. random variable.
  • Risk aversion. In utility theory this quantity characterizes how investor dislikes risk. More risk averse investor would less likely be involved in highly risky projects. For an utility function U the (Pratt) risk aversion is calculated as

    - U''(x) / U'(x).

    For exponential utility function

    U(x) = (1 - exp( - a * x )) / (1 - exp( - a )),

    risk aversion coincides with value of parameter a.

  • Risk measure. "Quantity" of risk contained in a given risk; a functional on the space of probability distributions. Examples are: expected utility, Value at risk, distorted probability. Classic problems of risk theory use variance and standard deviation as risk measures. Partial case of risk measure is risk price. Each risk measure generates a notion of certainty equivalent.
  • Risk price. A special case of risk measure. Risk price is used to calculate risk premium when transferring risk from one carrier to another. Examples are: insurance, options and other derivative instruments.
S T
U
  • Uncertainty. Lack of information about object of interest. For example there might be known the set of possible states X={x,y,z,...} of the object, but exact state is unknown. Uncertainty is called probabilistic if one knowns the probability distribution on X that describes probabilities of "visiting" specific states by the object.
  • Uniform distribution. A distribution of a random variable X on an interval [a,b] such that probability of falling X within subinterval [c,d] of [a,b] is proportional to its length. In multidimentional case the probability of falling a random vector X within a specific body is proportional to volume of a body. This distribution is sometimes called geometric. More.
  • Utility function. In utility theory this is a function U, prescribing a quantity named "utility" to wealth, outcomes of experiments, results of activities, etc. The latter are often reals, in which case U is a real function. It is often thought of as increasing concave function. In risk theory a derivative notion of expected utility is widely used.
V W
X
Y
Z


Sign guest book      View guest book