Binomial and Normal distributions used in business forecasting

Made By: Abhay Singh Roll No.50202 BBS I A

Business Statistics and Applications Term Paper

By Abhay Singh

STATISTICS TERM PAPER

Figure

What are theoretical distributions?

Theoretical distributions are a more scientific way of drawing inferences about the population characteristics. In the population, the value of the variable may be distributed according to some definite probability law which can be expressed mathematically and the corresponding probability distribution is known as the theoretical probability distribution. Such probability laws may be based on a prior considerations or a posteriori inferences. These distributions are based on expectations on the bass of previous experience. Theoretical distributions also enable us to fit a mathematical model or a function of the form y = p(x) to the given data. Refer to Figure on page 4

Discrete y

Probabilit
Binomial Distribution

Figure

In probability theory and statistics, the binomial distribution is the discrete probability distribution of the number of successes in a sequence of n independent yes/no experiments, each of which yields success with probability p. Such a success/failure experiment is also called a Bernoulli experiment or Bernoulli trial; when n = 1, the binomial distribution is a Bernoulli distribution. The binomial distribution is the basis for the popular binomial test of statistical significance. The binomial distribution is frequently used to model the number of successes in a sample of size n drawn with replacement from a population of size N. If the sampling is carried out without replacement, the draws are not independent and so the resulting distribution is a hypergeometric distribution, not a binomial one. However, for N much larger than n, the binomial distribution is a good approximation, and widely used.

Continuous Probabilit y

Cumulative Distribution Function

Probability Mass Function

Why is it important? The binomial distribution is widely used to test statistical probabilities and significance, and is a good way of visually detecting unexpected values. It is a useful tool in determining permutations, combinations, and probabilities, where the outcomes can be broken Binomial Distribution into two probabilities (p and q), where p and q are complementary (i.e., p + q = 1). For example, tossing a coin has only two possible outcomes, heads or tails. Each of these outcomes has a theoretical probability of 0.5. Using the binomial expansion, showing all possible outcomes and combinations, the probability is represented as follows:

(p + q)2 = p2 + 2pq +q2, or more simply, pp + 2pq + qq

If p is heads and q is tails, the theory shows there is only one way to get two heads (pp), two ways to get a head and a tail (2pq), and one way to get two tails (qq). Common uses of binomial distributions in business include quality control, public opinion surveys, medical research, and insurance problems. It can be applied to complex processes such as sampling items in factory production lines or to estimate percentage failure rates of products and components.

Quick facts To satisfy the requirements of binomial distribution, the event being studied must display certain characteristics: the number of trials or occurrences are fixed there are only two possible outcomes (heads/tails or win/lose, for example) all occurrences are independent of each other (tossing a head does not make it more or less likely you will get the same result next time) all outcomes have the same probability of success Binomial distribution is best applied in cases where the population size is at least 10 times the sample size, and not to simple random samples. To find probabilities from a binomial distribution, you can perform a manual calculation, but there are online calculators available, or you can use a binomial table or computer spreadsheet. The binomial distribution is sometimes called a Bernoulli experiment or trial. The binomial probability refers to the probability that a binomial experiment results in exactly x successes. In example above, we see that the binomial probability of getting exactly one head in two coin flips is 0.5. A cumulative binomial probability refers to the probability that the binomial random variable falls within a specified range (for example, is greater than or equal to a stated lower limit and less than or equal to a stated upper limit).

Uses in Business
1. Quality Control
In statistical quality control, the p-chart is a type of control chart used to monitor the proportion of nonconforming units in a sample, where the sample proportion nonconforming is defined as the ratio of the number of nonconforming units to the sample size, n. The p-chart only accommodates "pass"/"fail"-type inspection as determined by one or more go-no go gauges or tests, effectively applying the specifications to the data before they are plotted on the chart. Other types of control charts display the magnitude of the quality characteristic under study, making troubleshooting possible directly from those charts. The binomial distribution is the basis for the p-chart and requires the following assumptions [2]:267: The probability of nonconformity p is the same for each unit; Each unit is independent of its predecessors or successors;

The inspection procedure is same for each sample and is carried out consistently from sample to sample

A P-Chart

2. Public Opinion Survey When doing a public opinion poll before an election in a country, they usually approximate the hypergeometric distribution with a binomial distribution and then using the normal approximation: m +/-1.96*sqrt(m*(1-m)/n) to calculate a 95% confidence interval? m= mean n= number of people in the poll is it as simple as that? assuming party A gets 30% of the votes, and 2000 voted we get a 95% statistically significant interval of: 30 +/-1.96*sqrt(0.30*(0.7)/2000) However if we make a poll within a defined population of lets say 5000 people. and the number of people in the poll is 3000. party A gets 30% of the votes in the poll. What can we say about the total population of 5000. here we should use the hypergeometric distribution and cannot use the binomial approximation.

3. Medical Research
A binomial distribution can be used to describe the number of times an event will occur in a group of patients, a series of clinical trials, or any other sequence of observations. This event is a binary variable: It either occurs or it doesn't. For example, when patients are treated with a new drug they are either cured or not; when a coin is flipped, the result is either a head or tail. The binary outcome associated with each event is typically referred to as either a "success" or a "failure." In general, a binomial distribution is used to characterize the number of successes over a series of observations (or trials), where each observation is referred to as a "Bernoulli trial." In a series of n Bernoulli trials, the binomial distribution can be used to calculate the probability of obtaining k successful outcomes. If the variable X represents

the total number of successes in n trials, it can only take on a value from 0 to n. The binomial distribution can be used to calculate the probability of obtaining k successes in n trials is calculated as follows: where 0 less than or equal to p less than or equal to 1 is the probability of success, and n!= 1 2 3[.dotmath][.dotmath][.dotmath][.dotmath] (n2)(n1)n. The above formula assumes that the experiment consists of n identical trials that are independent from one another, and that there are only two possible outcomes for each trial (success or failure). The probability of success ( p) is also assumed to be the same in each of the trials. To further illustrate the application of the above formula, if a drug was developed that cured 30 percent of all patients, and it was administered to ten patients, the probability that exactly four patients would be cured is: Like other distributions, the binomial distribution can be described in terms of a mean and the spread, or variance, of values. The mean value of a binomial random variable X (i.e., the average number of successes in n trials) can be obtained by multiplying the number of trials by p (np). In the above example, the average number of persons cured in any group of 10 patients would thus be 3. The variance of a binomial distribution is np (1p). The variance is largest for p = 0.5, while it decreases as p approaches 0 or 1. Intuitively, this makes sense, since when p is very large or small nearly all the outcomes take on the same value. Returning to the example, a drug that cured every patient p would equal one, while for a drug that cured no one, p would equal zero. In contrast, if the drug was effective in curing only half of the population (p = 0.5) it would be more difficult to predict the outcome in any particular patient, and in this case the variability is relatively large. In studies of public health, the binomial distribution is used when a researcher is interested in the occurrence of an event rather than in its magnitude. For instance, smoking cessation interventions may choose to focus on whether a smoker quit smoking altogether, rather than evaluate daily reductions in the number of cigarettes smoked. The binomial distribution plays an important role in statistics, as it is likely the most frequently used distribution to describe discrete data.

4. Insurance Sector
There is similarity between insurance and game of chance and therefore understanding the concept of probability and its application to general insurance is of importance to us. When we toss a coin, we say that the probability of getting a head is . There are two

possibilities head or tail out of which one possibility i.e. head is favourable and therefore we say that probability of getting head is or 50%. This is the theoretical way of calculating probability called a priori i.e. prior to experience. In contrast, we can not deduce theoretically the probability of a car being stolen within the year. For handling problems of this nature, we need to have data about the total number of cars and the proportion that is stolen. There is another way of looking at probability of getting head in tossing of coin which is more relevant for our purpose. If we go on tossing the coin and note the number of heads coming and if ideally this tossing is continued for infinite number of times we will find that the proportion of head coming is . Every time we are tossing, we are in effect generating experience. The fact that probability involves long run concept is important in the general insurance contract. Further head & tail are mutually exclusive events. The idea of mutual exclusivity can apply for example to calculating the probability of an injured employee being male or female, injured or killed, damages being above or below certain level, etc. We may say that if the event is certain to happen the probability is one and if the happening is absolute impossibility the probability is zero. If the probability of happening a claim is one i.e. a certainty no insurance company will assume such a risk except perhaps by charging the premium which is more than the sum insured. If the happening is an impossibility i.e. probability is zero, nobody would like to insure it. Between these two extremes, lies the various risk that come for insurance. The higher the probability of claims happening, the higher should be the premium. Probability thus attaches a numerical value to our measurement of the likelihood of an event occurring. We shall now examine the law of large number and the concept of probability distribution. We shall also see how these probability distribution help us in estimating the number of claims that will be reported in future

during a given period of time and what will be the size of these claims. The law of large numbers in simple terms means that the larger the data, the more accurate will be the estimate made. In other words the larger the sample, the more accurate will be the estimates of the population parameters. In general insurance it would mean that the larger the past data about claims, the better will be the estimate of the prediction about claim frequency and size. It is assumed that the claim will occur in future as they have occurred in the past. What is a probability distribution? It is the listing of all possible values of a random variable along with their associated probabilities. For our purpose, the probability distribution can be considered to be a mathematical model which can describe the actual probability distribution. Of course, the actual probability estimated from the available data will rarely coincide with those generated by the theoretical distribution. But the law of large number says that it will tend closer & closer if we have sufficiently large database. Even if the data available is not extensive, we can make use of various
theoretical distributions to make meaningful inferences about the behaviour of data relating to a particular insurance portfolio. The fact that this theoretical distribution can be completely summarized by a small number of parameters is of great help. The shape of distribution is determined by its parameters. Parameters are numerical characteristics of population. If we have set of data relating to say claims size, we cannot make best use of them in their raw form. We may be interested to know about the average size of the claim. We have a whole set of measures called the measure of central tendency. Similarly to properly understand the significance of the data, it is essential to know the variability of data around the central tendency. In case the variance is too high, may be one has to decide about the required reinsurance support. Yet another aspect to properly understand the given set of data is the Skew ness aspect. The distribution may be very symmetric or it may be skewed having long trail to the right (positively skewed). Many of the distribution we encounter in general insurance is skew with long tail to the right. We have a measure of this skew ness which is zero for symmetric distribution. Positive for

positively skewed distribution and negative for negatively skewed distribution (long tail to the left). Again a knowledge and measure of flatness or peaked ness of a distribution is important for us. So we have what is called a measure of kurtosis. These aspects if known properly can help in better claims management. Fortunately for us, there are theoretical distribution models which approximate the existing claims data relating to various risk categories. The actuaries make use of these models. These provide methods of summarizing aspects of complexities. Some distributions are continuous in nature and may relate to claim size distribution and in the analysis of heterogeneity. The others relate to discrete variables and hence are helpful in studying the claim numbers distribution.

Normal Distribution In probability theory, the normal (or Gaussian) distribution is a continuous probability distribution that is often used as a first approximation to describe real-valued random variables that tend to cluster around a single mean value. The graph of the associated probability density function is "bell"-shaped, and is known as the Gaussian function or bell curve.

where parameter is the mean or expectation (location of the peak) and 2 is the variance, the mean of the squared deviation, (a "measure" of the width of the distribution). is the standard deviation. The distribution with = 0 and 2 = 1 is called the standard normal. The normal distribution is considered the most prominent probability distribution in statistics. There are several reasons for this: First, the normal distribution is very tractable analytically, that is, a large number of results involving this distribution can be derived in explicit form. Second, the normal distribution arises as the outcome of the central limit theorem, which states that under mild conditions the sum of a large number of random variables is distributed approximately normally. Finally, the "bell" shape of the normal distribution makes it a convenient choice for modelling a large variety of random variables encountered in practice. For this reason, the normal distribution is commonly encountered in practice, and is used throughout statistics, natural sciences, and social sciences as a simple model for complex phenomena. For example, the observational error in an experiment is usually assumed to follow a normal distribution, and the propagation of uncertainty is computed using this assumption. Note that a normally-distributed variable has a symmetric distribution about its mean.

Quantities that grow exponentially, such as prices, incomes or populations, are often skewed to the right, and hence may be better described by other distributions, such as the log-normal distribution or Pareto distribution. In addition, the probability of seeing a normally-distributed value that is far (i.e. more than a few standard deviations) from the mean drops off extremely rapidly. As a result, statistical inference using a normal distribution is not robust to the presence of outliers(data that is unexpectedly far from the mean, due to exceptional circumstances, observational error, etc.). When outliers are expected, data may be better described using a heavytailed distribution such as the Student's t-distribution.

Probability Density Function

Cumulative Descriptive Function Uses 1. Modern Portfolio Theory

Modern portfolio theory (MPT)or portfolio theorywas introduced by Harry Markowitz with his paper "Portfolio Selection," which appeared in the 1952 Journal of Finance. Thirty-eight years later, he shared a Nobel Prize with Merton Miller and William Sharpefor what has become a broad theory for portfolio selection.

Prior to Markowitz's work, investors focused on assessing the risks and rewards of individualsecurities in constructing their portfolios. Standard investment advice was to identify those securities that offered the best opportunities for gain with the least risk and then construct a portfolio from these. Following this advice, an investor might conclude that railroad stocks all offered good risk-reward characteristics and compile a portfolio entirely from these. Intuitively, this would be foolish. Markowitz formalized this intuition. Detailing a mathematics of diversification, he proposed that investors focus on selecting portfolios based on their overall risk-reward characteristics instead of merely compiling portfolios from securities that each individually have attractive risk-reward characteristics. In a nutshell, inventors should select portfolios not individual securities.

If we treat single-period returns for various securities as random variables, we can assign them expected values, standard deviations and correlations. Based on these, we can calculate the expected return and volatility of any portfolio constructed with those securities. We may treat volatility and expected return as proxy's for risk and reward. Out of the entire universe of possible portfolios, certain ones will optimally balance risk and reward. These comprise what Markowitz called an efficient frontier of portfolios. An investor should select a portfolio that lies on the efficient frontier. James Tobin (1958) expanded on Markowitz's work by adding a risk-free asset to the analysis. This made it possible to leverage or deleverage portfolios on the efficient frontier. This lead to the notions of a superefficient portfolio and the capital market line. Through leverage, portfolios on the capital market line are able to outperform portfolio on the efficient frontier.

Sharpe (1964) formalized the capital asset pricing model(CAPM). This makes strong assumptions that lead to interesting conclusions. Not only does the market portfolio sit on the efficient frontier, but it is actually Tobin's super-efficient portfolio. According to CAPM, all investors should hold the market portfolio, leveraged or de-leveraged with positions in the risk-free asset. CAPM also introduced beta and relates an asset's expected return to its beta. Portfolio theory provides a broad context for understanding the interactions of systematic risk and reward. It has profoundly shaped how institutional portfolios are managed, and motivated the use of passive

investment management techniques. The mathematics of portfolio theory is used extensively in financial risk management and was a theoretical precursor for today's value-at-risk measures.

2. Human resource Management

A disadvantage shared by all employee-comparison systems is that of employee comparability. This has two aspects. The first has been mentioned: are the jobs sufficiently similar? The second is whether employees are rated on the same criteria. It is likely that one employee rates high for one reason and another rates low for an entirely different reason. Another disadvantage is that raters do not always have sufficient knowledge of the people being rated. Normally the immediate supervisor has this knowledge, but in large ranking systems, supervisors two and three levels removed often have to do the rating. The very size of units also poses a problem. The larger the number of employees to be ranked, the harder it is to do so; on the other hand, the larger the number in the group, the more logical it is that there is a normal distribution. This brings up one last problem. If the manager knows that some employees must be rated below average, he or she will start thinking of those employees that way. This leads to a self-fulfilling prophecy: the manager now treats them as if they cannot do well, and they respond by not doing well.

Forecasting
The most comprehensive methodology for comparing forecast data and demand data is to graph the cumulative results for a given period which include upper and lower control limits (calculated from historic demand based on a normal distribution curve). The following example shows how easy it is to quickly identify when the demand is in control and when the demand is out of control.

This graph shows a period of meeting the plan (Note A) and then a period when the actual revenues start to deviate from the forecast (Note B). At this point, the demand is still within the control parameters of "normal" demand and a cautionary watch may be put in place, although no action is required. However, it is clear at Note C the lower control limit has been exceeded and the "normal" expected demand is not being met. This is the time for action and the process to gain a complete understanding of the error should be invoked. Supplemental charts will be necessary to analyze what is causing the deviation. These charts are similar to the one shown above, but with a separate breakdown for units, average selling price, product types, sales channels, customer and sales agent. A possible result is finding that the graph represents a normal trend in the business with no corrective actions necessary. It is also possible that the total revenue versus forecast may be in sync (note A), however mismatches may exist in the unit, average selling price, customer or product type mixtures. In each case, this information would not be known unless this type of analysis were available as well as being in place for some time in order to understand the long term trends as well.

Applications:
1. Using the Normal Distribution to Determine the Lower 10% Limit of Delivery Times
A pizza deliveryman's delivery time is normally distributed with a mean of 20 minutes and a standard deviation of 4 minutes. What delivery time will be beaten by only 10% of all deliveries?

Problem Parameter Outline

Population Mean = = "mu" = 20 minute Population Standard Deviation =

= "sigma" = 4 minutes

x=?

Probability that (delivery time x) = 10% = 0.10 Delivery time is Normally distributed Normal curve is not standardized ( 0, 1)

Problem Solving Steps

We know that Delivery Time data is Normally distributed and can therefore be mapped on the Normal curve. We are trying to determine Delivery Time will be lower than 90% of all Delivery Times. This probability corresponds to the x value at which 90% of area under the Normal curve has a greater value and is to the right this x value. This x value must therefore be in the left tail of the Normal curve. If we know that 90% of the area under the Normal curve is to the right of this x value (this is illustrated in the graph below), then we know that 40% of the total area under the Normal curve is between this x value and the mean. The remaining 50% of the area under the Normal curve makes up the half the Normal curve that is on the opposite side (the right side) of the mean. If we know that 40% of the area under the Normal curve is between this x value and the mean, we can use the Z Score Chart to determine how many standard deviations this x value is from the mean. The Z Score Chart belowshows this x value to be 1.28 standard deviations from the mean. If we know how many standard deviations this x value is from the mean, we can use the following formula to calculate the x value, as follows:

z=(x-)/ x=z*+ x=
(-1.28) * 4 + 20 = 14.87

The delivery time of 14.87 minutes is faster (smaller) than 90% of all delivery times. This is illustrated in the graph below. Answer: The delivery time of 14.87 minutes is faster (smaller) than 90% of all delivery times.

2. Finding the probability of a certain type of package passing down a conveyor belt if the probability of that type of package passing by is known.
Problem: A conveyor belt brings packages to a truck to loaded. The packages are either black or white. The probablity that a package is black is 40%. What is the probability that out of the next 10 packages, at least 2 are black and 2 are white? The only possibility of at least two packages being white and 2 black would occur if the number of black packages equaled 0, 1, 9, or 10. The probability of at least 2 packages being black and 2 white would therefore equal 1 minus the probability that the number of black packages equals 0, 1, 9, or 10. Probability of Success (Black Package) = p = 0.40 Probability of No Success (White Package) = q = 1 - p = 0.60 Number of Trials (Transactions) = n = 10 Exact Number of Successes = k = 0, 1, 9, 10

This problem uses Binomial formulas (Sample Occurrence Formulas) because what is being measured is the mean number of occurrences from a large number of individual samples that each have only two possible outcomes. Probability of at least 2 Black and 2 White Packages in 10 = 1 - [ Pr(X=0) + Pr(X=1) + Pr(9) + Pr(10) ] PR (X = 0) = f(k; n, p) = n! / [ k! * ( n - k )! ] * pk * q(n-k) PR (X = 0) = f(0; 10, 0.40) = 0! / [ 10! * ( 10 - 0 )! ] * (0.40)\ 0 * (0.60)(10-0) PR (X = 0) = 0.0060 = 0.60% PR (X = 1) = f(k; n, p) = n! / [ k! * ( n - k )! ] * pk * q(n-k) PR (X = 1) = f(1; 10, 0.40) = 1! / [ 10! * ( 10 - 1 )! ] * (0.40)\ 1 * (0.60)(10-1) PR (X = 1) = 0.040 = 4.0% PR (X = 9) = f(k; n, p) = n! / [ k! * ( n - k )! ] * pk * q(n-k)

PR (X = 9) = f(9; 10, 0.40) = 9! / [ 10! * ( 10 - 9 )! ] * (0.40)\ 9 * (0.60)(10-9) PR (X = 9) = 0.002 = 0.2% PR (X = 10) = f(k; n, p) = n! / [ k! * ( n - k )! ] * pk * q(n-k) PR (X = 10) = f(10; 10, 0.40) = 10! / [ 10! * ( 10 - 10 )! ] * (0.40) 10 * (0.60)(10-10) PR (X = 10) = 0.0 = 0.0% 1 - [ Pr(X=0) + Pr(X=1) + Pr(9) + Pr(10) ] = 1 - [ 0.006 + 0.040 + 0.002 + 0.0 ] = 1 - 0.048 = 0.952 = 95.2% There is a 95.2% probability at at least 2 packages will be black and 2 packages will be white out of the next 10 packages if the probability of a package being black is 40%.

I
Insurance, 67

T
the binomial distribution. See Page 2 Theoretical distributions, 2