Term 1 · Module 2 of 4

Probability and Random Variables

Business Statistics for Entrepreneurs I

Practise this module in Claude, ChatGPT or Gemini

Set it up free
Mind map

Probability: experiments, events, and basic rules

Probability measures how likely an event is, on a scale from 00 (impossible) to 11 (certain). A probability of 0.50.5 represents equal likelihood of occurrence and non-occurrence. Probabilities can neither be negative nor exceed 11.

A random experiment is a procedure with more than one possible outcome. Its sample space SS is the set of all possible outcomes; each individual outcome is a sample point. An event is a subset of SS. Its probability is the sum of the probabilities of its sample points. The complement AcA^c contains outcomes in SS that are not in AA.

Movie-revenue example

Consider the joint outcomes for KGF Chapter 2 revenue of ₹3, ₹4, ₹5, or ₹6 lakh and Kantara revenue of ₹4, ₹5, or ₹6 lakh. This experiment has 4×3=124\times3=12 possible outcome pairs. Let AA mean KGF is a hit and BB mean Kantara is a hit. Taking hit to mean revenue of at least ₹5 lakh, the source identifies outcomes 7–12 for AA, outcomes 2, 3, 5, 6, 8, 9, 11, 12 for BB, and outcomes 8, 9, 11, 12 for both.

The historical outcome probabilities sum to 11. Adding the relevant outcome probabilities gives P(A)=0.34P(A)=0.34 and P(B)=0.60P(B)=0.60. Thus:

P(Ac)=1−P(A)=1−0.34=0.66,P(Bc)=1−P(B)=1−0.60=0.40.P(A^c)=1-P(A)=1-0.34=0.66,\qquad P(B^c)=1-P(B)=1-0.60=0.40.

The union A∪BA\cup B means AA or BB or both; the intersection A∩BA\cap B means both. The example gives P(A∩B)=0.08+0.02+0.06+0.02=0.18P(A\cap B)=0.08+0.02+0.06+0.02=0.18. Adding P(A)P(A) and P(B)P(B) counts the overlap twice, so subtract it once:

P(A∪B)=P(A)+P(B)−P(A∩B)=0.34+0.60−0.18=0.76.P(A\cup B)=P(A)+P(B)-P(A\cap B)=0.34+0.60-0.18=0.76.

Mutually exclusive events have no outcomes in common, so P(A∩B)=0P(A\cap B)=0 and P(A∪B)=P(A)+P(B)P(A\cup B)=P(A)+P(B). A single coin toss cannot produce both heads and tails.

Conditional probability, Bayes’ theorem, and independence

Conditional probability updates the probability of an event after learning that another event occurred. For P(B)>0P(B)>0:

P(A∣B)=P(A∩B)P(B),P(B∣A)=P(A∩B)P(A).P(A\mid B)=\frac{P(A\cap B)}{P(B)},\qquad P(B\mid A)=\frac{P(A\cap B)}{P(A)}.

Equivalently, the multiplication rule is P(A∩B)=P(A∣B)P(B)=P(B∣A)P(A)P(A\cap B)=P(A\mid B)P(B)=P(B\mid A)P(A). In the movie example, P(A∣B)=0.18/0.60=0.30P(A\mid B)=0.18/0.60=0.30: given Kantara is a hit, the probability KGF is a hit is 30%30\%. In the reverse direction, P(B∣A)=0.18/0.34≈0.53P(B\mid A)=0.18/0.34\approx0.53: among cases where KGF is a hit, the probability Kantara is a hit is about 53%53\%. A joint-probability table places joint probabilities in interior cells and marginal probabilities on the margins; divide a joint cell by the conditioning event’s marginal to find a conditional probability.

Bayes’ theorem expresses the same update by reversing the conditioning:

P(A∣B)=P(B∣A)P(A)P(B).P(A\mid B)=\frac{P(B\mid A)P(A)}{P(B)}.

Here P(A)P(A) is a prior probability and P(A∣B)P(A\mid B) is the posterior after observing BB. For instance, the movie example’s posterior P(B∣A)=0.53P(B\mid A)=0.53 is below its prior P(B)=0.60P(B)=0.60.

Events AA and BB are independent when knowing one occurred does not change the probability of the other: P(B∣A)=P(B)P(B\mid A)=P(B) (equivalently, P(A∣B)=P(A)P(A\mid B)=P(A)). For events with defined probabilities, this is equivalent to:

P(A∩B)=P(A)P(B).P(A\cap B)=P(A)P(B).

The movie events are not independent because 0.18≠(0.34)(0.60)=0.2040.18\ne(0.34)(0.60)=0.204. Mutually exclusive events with nonzero probabilities cannot also be independent: occurrence of one rules out the other.

The law of total probability partitions an event across mutually exclusive and collectively exhaustive cases. For a partition B1,…,BnB_1,\ldots,B_n of SS:

P(A)=∑i=1nP(A∩Bi)=∑i=1nP(A∣Bi)P(Bi).P(A)=\sum_{i=1}^{n}P(A\cap B_i)=\sum_{i=1}^{n}P(A\mid B_i)P(B_i).

For the two-case partition B,BcB,B^c, P(A)=P(A∩B)+P(A∩Bc)P(A)=P(A\cap B)+P(A\cap B^c). In the movie example, partitioning by Kantara revenue of ₹4, ₹5, or ₹6 lakh gives the same P(A)=0.34P(A)=0.34 by summing the three corresponding joint probabilities.

Joint table: work experience and salary

In a sample of 103103 respondents, work experience is grouped as at most two years or more than two years; salary as at most ₹25 lakh or above ₹25 lakh. The four cell counts are respectively 34,13,22,3434,13,22,34 (rows: at most two years, more than two years; columns: salary at most ₹25 lakh, above ₹25 lakh). Margins are 47,5647,56 for experience and 56,4756,47 for salary.

  • More than two years’ experience: 56/103≈0.5456/103\approx0.54.
  • Salary above ₹25 lakh: 47/103≈0.4647/103\approx0.46.
  • At most two years’ experience and salary above ₹25 lakh: 13/103≈0.1313/103\approx0.13.
  • More than two years’ experience and salary at most ₹25 lakh: 22/103≈0.2122/103\approx0.21.
  • Given more than two years’ experience, salary above ₹25 lakh: 34/56≈0.6134/56\approx0.61.
  • Given salary at most ₹25 lakh, more than two years’ experience: 22/56≈0.3922/56\approx0.39.

Key takeaways

  • An event is a set of outcomes; sum the probabilities of its outcomes.
  • Union includes overlap; subtract the intersection once.
  • Conditional probability divides the joint probability by the conditioning event’s probability.
  • Independence is tested by a conditional probability or the product rule.
  • A partition lets total probability be computed as a sum over exhaustive cases.

Added probability examples: worked applications

FinTech company locations

Among 150150 companies, one is selected at random. Of these, 2828 are in Maharashtra, 2929 in Karnataka, and 2626 in Haryana. Since locations are mutually exclusive:

P(M)=28150≈0.187,P(K)=29150≈0.193,P(H)=26150≈0.173.P(M)=\frac{28}{150}\approx0.187,\quad P(K)=\frac{29}{150}\approx0.193,\quad P(H)=\frac{26}{150}\approx0.173.

The chance of a company being in any of those three states is:

P(M∪K∪H)=28+29+26150=83150≈0.553.P(M\cup K\cup H)=\frac{28+29+26}{150}=\frac{83}{150}\approx0.553.

The complement of Karnataka is:

P(Kc)=1−29150=121150≈0.807.P(K^c)=1-\frac{29}{150}=\frac{121}{150}\approx0.807.

Teenagers’ expected age of financial independence

Among 9696 surveyed teenagers, counts for expected independence at ages 21–24, 25–27, 28–32, and 33 or older are 19,38,22,1719,38,22,17. Thus the category probabilities are:

1996≈0.198,3896≈0.396,2296≈0.229,1796≈0.177.\frac{19}{96}\approx0.198,\quad\frac{38}{96}\approx0.396,\quad\frac{22}{96}\approx0.229,\quad\frac{17}{96}\approx0.177.

Before age 28 combines the first two categories:

P(before 28)=19+3896=5796≈0.594.P(\text{before 28})=\frac{19+38}{96}=\frac{57}{96}\approx0.594.

After age 32 is the last category: 17/96≈0.17717/96\approx0.177. The transcript interprets the survey as roughly 60%60\% expecting independence before 28 and 82%82\% by age 32; it characterizes those expectations as seeming reasonable, without supplying an external benchmark.

Mutual-fund ratings and type

For 2525 funds, 1616 are domestic-equity, 1313 are rated three stars or below, seven domestic funds have four stars, and two domestic funds have five stars. Therefore 25−13=1225-13=12 funds have four or five stars. The number of domestic funds rated one to three stars is 16−(7+2)=716-(7+2)=7; non-domestic funds rated four or five stars number 12−7−2=312-7-2=3.

P(domestic)=1625=0.64,P(4 or 5 stars)=1225=0.48.P(\text{domestic})=\frac{16}{25}=0.64,\quad P(4\text{ or }5\text{ stars})=\frac{12}{25}=0.48.

P(domestic and 4 or 5 stars)=7+225=925=0.36.P(\text{domestic and 4 or 5 stars})=\frac{7+2}{25}=\frac{9}{25}=0.36.

Using the disjoint groups domestic/low-rated, domestic/high-rated, and non-domestic/high-rated:

P(domestic or 4/5 stars)=7+9+325=1925=0.76.P(\text{domestic or 4/5 stars})=\frac{7+9+3}{25}=\frac{19}{25}=0.76.

Student loans and prior college degree

Let YY denote prior college degree, NN no prior degree, SS satisfactory repayment, and DD delinquency. The joint table gives P(Y∩S)=0.20P(Y\cap S)=0.20, P(Y∩D)=0.18P(Y\cap D)=0.18, P(N∩S)=0.24P(N\cap S)=0.24, and P(N∩D)=0.38P(N\cap D)=0.38. Hence P(Y)=0.20+0.18=0.38P(Y)=0.20+0.18=0.38 and P(N)=0.24+0.38=0.62P(N)=0.24+0.38=0.62.

P(D∣Y)=0.180.38≈0.47,P(D∣N)=0.380.62≈0.61.P(D\mid Y)=\frac{0.18}{0.38}\approx0.47,\qquad P(D\mid N)=\frac{0.38}{0.62}\approx0.61.

The joint probability of no degree and satisfactory repayment is P(N∩S)=0.24P(N\cap S)=0.24. The observed delinquency rate is higher among recipients without a prior degree in this table; this association alone does not establish causation or a sound lending criterion.

Missed payment and credit-card default

The prior default probability is P(D)=0.08P(D)=0.08, so P(Dc)=0.92P(D^c)=0.92. The transcript states that P(P∣Dc)=0.20P(P\mid D^c)=0.20 in the setup, but later uses 0.250.25 in the calculation. Its worked calculation therefore uses 0.250.25: P(P∩Dc)=0.25(0.92)=0.23P(P\cap D^c)=0.25(0.92)=0.23 and P(Pc∩Dc)=0.92−0.23=0.69P(P^c\cap D^c)=0.92-0.23=0.69. Since defaults always miss a payment, P(D∩P)=0.08P(D\cap P)=0.08 and P(D∩Pc)=0P(D\cap P^c)=0; thus P(P)=0.08+0.23=0.31P(P)=0.08+0.23=0.31.

P(D∣P)=P(D∩P)P(P)=0.080.31≈0.258.P(D\mid P)=\frac{P(D\cap P)}{P(P)}=\frac{0.08}{0.31}\approx0.258.

This exceeds the stated recall threshold of 0.250.25, so under that rule the worked example says to recall the card after a missed payment. The source’s 0.200.20 versus 0.250.25 inconsistency is unresolved; the displayed posterior uses 0.250.25.

Household education and income

The study includes 721721 households, with education categories (primary, secondary, college, advanced) and income categories (₹4–6 lakh, ₹6–8 lakh, ₹8–12 lakh, ₹12 lakh or more). Each joint probability is a cell count divided by 721721; summing across a row or column gives its marginal probability. The narrated table supplies these calculations:

  • College or advanced degree: marginal probability 0.170.17.
  • Income of at least ₹12 lakh given primary education: 0.05/0.50=0.100.05/0.50=0.10.
  • Income ₹4–6 lakh: marginal probability 0.200.20.
  • Income ₹8–12 lakh given college education: 0.05/0.14≈0.360.05/0.14\approx0.36 (the transcript says roughly 0.330.33, an arithmetic inconsistency).
  • The primary/₹4–6 lakh joint cell is 0.150.15, while the product of its stated marginals is (0.20)(0.50)=0.10(0.20)(0.50)=0.10. Since they differ, education and income are not independent.

Key takeaways

  • For equally likely selections from a finite list, probability is count divided by total.
  • Combine disjoint categories by adding their probabilities; use complements for “not.”
  • Joint tables provide joint, marginal, and conditional probabilities from the same counts.
  • Check arithmetic against the stated inputs; preserve unresolved source inconsistencies explicitly.

Random variables: meaning and types

A random variable assigns a numerical value to each outcome of a random experiment. The capital letter XX names the variable; lowercase xx denotes a value it can take. A coin toss may be coded X=0X=0 for heads and X=1X=1 for tails. A biased coin can have P(X=0)=pP(X=0)=p and P(X=1)=1−pP(X=1)=1-p. Other coin-related variables include the number of heads in ten tosses (values 00 through 1010) and the number of tosses until the first head (positive integer values).

A die outcome can be X∈{1,2,3,4,5,6}X\in\{1,2,3,4,5,6\}. For a fair die, each value has probability 1/61/6. For the example biased die, probabilities are 1/31/3 at 11, 1/61/6 at 2,3,42,3,4, and 1/121/12 at 5,65,6; these sum to one.

Type Possible values Business or operational examples
Discrete random variable Countable values, often integers Orders from five calls: 00–55; defective radios in a lot of 50: 00–5050; customers arriving in a day.
Continuous random variable Any value in a range, including fractions Time between bank customers; actual fluid in a nominal 12-ounce can, between 11.5 and 12.5 ounces; reaction temperature between 100°C and 150°C.

One setting can yield both types: number of bank arrivals is discrete, while time between arrivals is continuous.

Discrete distributions: PMF, CDF, expectation, and spread

The probability mass function (PMF) f(x)=P(X=x)f(x)=P(X=x) assigns probability mass to each possible value. Each mass is nonnegative and all masses sum to one. The cumulative distribution function (CDF) is F(x)=P(X≤x)F(x)=P(X\le x); the tail probability is Fˉ(x)=P(X>x)=1−F(x)\bar F(x)=P(X>x)=1-F(x).

For a discrete random variable:

E[X]=∑xxf(x),Var⁡(X)=∑x(x−E[X])2f(x),σX=Var⁡(X).E[X]=\sum_x x f(x),\qquad \operatorname{Var}(X)=\sum_x (x-E[X])^2f(x),\qquad \sigma_X=\sqrt{\operatorname{Var}(X)}.

Expectation is a probability-weighted center, not necessarily an attainable outcome. Variance is the probability-weighted squared distance from that center; standard deviation returns spread to the variable’s units.

Fair die

For a fair die, f(x)=1/6f(x)=1/6 for x=1,…,6x=1,\ldots,6. Therefore F(5)=5/6F(5)=5/6, and:

E[X]=∑x=16x6=216=3.5.E[X]=\sum_{x=1}^{6}\frac{x}{6}=\frac{21}{6}=3.5.

Var⁡(X)=(1−3.5)2+⋯+(6−3.5)26=3512≈2.92,σX=35/12≈1.71.\operatorname{Var}(X)=\frac{(1-3.5)^2+\cdots+(6-3.5)^2}{6}=\frac{35}{12}\approx2.92,\qquad \sigma_X=\sqrt{35/12}\approx1.71.

Even outcome: P(X∈{2,4,6})=3/6=1/2P(X\in\{2,4,6\})=3/6=1/2. Between 2 and 5 inclusive: 4/6=2/34/6=2/3. Greater than 3: 1−F(3)=1−3/6=1/21-F(3)=1-3/6=1/2.

Biased die

With f(1)=1/3f(1)=1/3, f(2)=f(3)=f(4)=1/6f(2)=f(3)=f(4)=1/6, and f(5)=f(6)=1/12f(5)=f(6)=1/12:

P(X=2)=16,F(5)=13+16+16+16+112=1112.P(X=2)=\frac16,\quad F(5)=\frac13+\frac16+\frac16+\frac16+\frac1{12}=\frac{11}{12}.

E[X]=1(13)+(2+3+4)(16)+(5+6)(112)=2.75.E[X]=1\left(\frac13\right)+(2+3+4)\left(\frac16\right)+(5+6)\left(\frac1{12}\right)=2.75.

Var⁡(X)=∑x=16(x−2.75)2f(x)≈2.69,σX=2.69≈1.64.\operatorname{Var}(X)=\sum_{x=1}^{6}(x-2.75)^2f(x)\approx2.69,\qquad \sigma_X=\sqrt{2.69}\approx1.64.

Even outcome: 1/6+1/6+1/12=5/121/6+1/6+1/12=5/12. Values 2 through 5 inclusive: 1/6+1/6+1/6+1/12=7/121/6+1/6+1/6+1/12=7/12. Greater than 3: 1/6+1/12+1/12=1/31/6+1/12+1/12=1/3, also 1−F(3)=1−2/3=1/31-F(3)=1-2/3=1/3.

Estimating a PMF from observed frequencies

When probabilities are not supplied, observed relative frequencies estimate them: f(x)=number of observations with value x/total observationsf(x)=\text{number of observations with value }x/\text{total observations}. For a pump factory’s 200 days, defective-pump counts 0,1,2,3,40,1,2,3,4 occurred 80,50,30,20,2080,50,30,20,20 times, giving probabilities 0.40,0.25,0.15,0.10,0.100.40,0.25,0.15,0.10,0.10. No day had more than four defects.

P(X=2)=0.15,P(X≤3)=0.40+0.25+0.15+0.10=0.90.P(X=2)=0.15,\quad P(X\le3)=0.40+0.25+0.15+0.10=0.90.

E[X]=0(0.40)+1(0.25)+2(0.15)+3(0.10)+4(0.10)=1.2.E[X]=0(0.40)+1(0.25)+2(0.15)+3(0.10)+4(0.10)=1.2.

The lecture reports variance 1.661.66 and standard deviation 1.291.29.

For late takeoffs in the 9–11 p.m. slot over 30 days, the transcript reports f(0)=3/30=0.10f(0)=3/30=0.10, f(1)=6/30=0.20f(1)=6/30=0.20, and the remaining displayed distribution only by resulting calculations. It gives P(X≥2)=1−f(0)−f(1)=0.70P(X\ge2)=1-f(0)-f(1)=0.70, P(X≤3)=0.80P(X\le3)=0.80, mean E[X]=2.20E[X]=2.20, variance 1.561.56, and standard deviation 1.56≈1.25\sqrt{1.56}\approx1.25. Translate the question into a set of values, then add their probabilities; for an upper tail, a complement may be quicker.

Key takeaways

  • PMF gives point probabilities; CDF accumulates through xx; tail probability is 1−F(x)1-F(x).
  • Expected value is a weighted center and can lie between possible outcomes.
  • Variance weights squared deviations by their probabilities; standard deviation is its square root.
  • Relative frequencies provide empirical probability estimates when a PMF is unknown.

Linear combinations of random variables

A linear combination Y=∑i=1naiXiY=\sum_{i=1}^n a_iX_i combines random inputs using fixed weights. Examples include material plus labor cost, store-level sales weighted by profit per unit, total expense across categories, and average sales commissions.

For constants a,ba,b:

Y=aX+b⟹E[Y]=aE[X]+b,Var⁡(Y)=a2Var⁡(X).Y=aX+b\quad\Longrightarrow\quad E[Y]=aE[X]+b,\qquad \operatorname{Var}(Y)=a^2\operatorname{Var}(X).

Shifting by bb moves the center but does not change spread; scaling by aa scales standard deviation by ∣a∣|a| and variance by a2a^2.

Unit conversions

For a die, E[X]=3.5E[X]=3.5 and Var⁡(X)=35/12\operatorname{Var}(X)=35/12. Transforming to Y=100X+10Y=100X+10 gives E[Y]=100(3.5)+10=360E[Y]=100(3.5)+10=360 and Var⁡(Y)=1002(35/12)\operatorname{Var}(Y)=100^2(35/12).

For frying temperature, Celsius XX has mean 160160 and standard deviation 55, so variance 2525. Fahrenheit Y=(9/5)X+32Y=(9/5)X+32:

E[Y]=95(160)+32=320,Var⁡(Y)=(95)2(25)=81,σY=9°F.E[Y]=\frac95(160)+32=320,\qquad \operatorname{Var}(Y)=\left(\frac95\right)^2(25)=81,\qquad \sigma_Y=9\text{°F}.

For Nariman Point property prices, rupee price XX has mean ₹36,000 per square foot and standard deviation ₹9,600. At ₹80 per dollar, Y=X/80Y=X/80:

E[Y]=36,00080=450 USD/ft2,Var⁡(Y)=9,6002802=14,400,σY=9,60080=120 USD/ft2.E[Y]=\frac{36{,}000}{80}=450\text{ USD/ft}^2,\quad \operatorname{Var}(Y)=\frac{9{,}600^2}{80^2}=14{,}400,\quad \sigma_Y=\frac{9{,}600}{80}=120\text{ USD/ft}^2.

Sums, differences, and covariance

For two variables, expectation is additive regardless of dependence. Variance includes a covariance term:

E[X1+X2]=E[X1]+E[X2],E[X_1+X_2]=E[X_1]+E[X_2],

Var⁡(X1+X2)=Var⁡(X1)+Var⁡(X2)+2ρ12σ1σ2.\operatorname{Var}(X_1+X_2)=\operatorname{Var}(X_1)+\operatorname{Var}(X_2)+2\rho_{12}\sigma_1\sigma_2.

For a difference, the covariance term changes sign:

Var⁡(X1−X2)=Var⁡(X1)+Var⁡(X2)−2ρ12σ1σ2.\operatorname{Var}(X_1-X_2)=\operatorname{Var}(X_1)+\operatorname{Var}(X_2)-2\rho_{12}\sigma_1\sigma_2.

If variables are independent, their covariance is zero, so variances add for either sum or difference. For independent X1,X2X_1,X_2 each with mean μ\mu and variance σ2\sigma^2, both X1+X2X_1+X_2 and X1−X2X_1-X_2 have variance 2σ22\sigma^2; their means are 2μ2\mu and 00 respectively.

For Y=∑iaiXiY=\sum_i a_iX_i, expectation is always ∑iaiE[Xi]\sum_i a_iE[X_i]. The variance is:

Var⁡(Y)=∑iai2Var⁡(Xi)+2∑i<jaiajCov⁡(Xi,Xj).\operatorname{Var}(Y)=\sum_i a_i^2\operatorname{Var}(X_i)+2\sum_{i<j}a_ia_j\operatorname{Cov}(X_i,X_j).

Since Cov⁡(Xi,Xj)=ρijσiσj\operatorname{Cov}(X_i,X_j)=\rho_{ij}\sigma_i\sigma_j, independence makes every pairwise term zero. For two stores with independent daily sales, means ₹7,000 and ₹8,000 and standard deviations ₹1,500 and ₹2,000:

E[X+Y]=15,000,SD⁡(X+Y)=1,5002+2,0002=2,500.E[X+Y]=15{,}000,\quad \operatorname{SD}(X+Y)=\sqrt{1{,}500^2+2{,}000^2}=2{,}500.

For the difference X−YX-Y, the mean is 7,000−8,000=−Rs. 1,0007{,}000-8{,}000=-\text{Rs. }1{,}000 and standard deviation remains ₹2,500.

Portfolio return: weights and diversification

Siemens returns XX have mean 3%3\% and standard deviation 1.5%1.5\%; Trent returns YY have mean 5%5\% and standard deviation 3%3\%. Their correlation is −0.6-0.6. With fraction ww in Siemens and 1−w1-w in Trent:

P=wX+(1−w)Y,P=wX+(1-w)Y,

E[P]=wE[X]+(1−w)E[Y],E[P]=wE[X]+(1-w)E[Y],

Var⁡(P)=w2σX2+(1−w)2σY2+2w(1−w)ρXYσXσY.\operatorname{Var}(P)=w^2\sigma_X^2+(1-w)^2\sigma_Y^2+2w(1-w)\rho_{XY}\sigma_X\sigma_Y.

At w=0.5w=0.5, expected return is 0.5(3%)+0.5(5%)=4%0.5(3\%)+0.5(5\%)=4\%. Variance in percentage-point-squared units is:

0.25(1.52)+0.25(32)+2(0.5)(0.5)(−0.6)(1.5)(3)=1.4625,0.25(1.5^2)+0.25(3^2)+2(0.5)(0.5)(-0.6)(1.5)(3)=1.4625,

so standard deviation is 1.4625≈1.21%\sqrt{1.4625}\approx1.21\%. Negative correlation reduces portfolio risk. The transcript’s comparison says a 90/10 Siemens/Trent portfolio has expected return 3.2%3.2\% and standard deviation 1.19%1.19\%; at roughly the same risk, the 50/50 portfolio has higher return and therefore dominates it. There is no single best split without a risk/return preference; Pareto optimal portfolios are undominated choices offering different trade-offs.

Sums and averages of independent, identically distributed variables

For independent variables with common mean μ\mu and variance σ2\sigma^2, their sum S=∑i=1nXiS=\sum_{i=1}^nX_i has:

E[S]=nμ,Var⁡(S)=nσ2,SD⁡(S)=n σ.E[S]=n\mu,\qquad \operatorname{Var}(S)=n\sigma^2,\qquad \operatorname{SD}(S)=\sqrt n\,\sigma.

If 81 passengers’ luggage weights are independent with mean 30 kg and standard deviation 3 kg, total weight has mean 81(30)=2,43081(30)=2{,}430 kg, variance 81(9)81(9), and standard deviation 729=27\sqrt{729}=27 kg. Its coefficient of variation, standard deviation divided by mean, is 27/2,430≈0.011=1.1%27/2{,}430\approx0.011=1.1\% (reported as roughly 1%1\%).

The sample average Xˉ=(1/n)∑iXi\bar X=(1/n)\sum_iX_i is itself a random variable. Under the same assumptions:

E[Xˉ]=μ,Var⁡(Xˉ)=σ2n,SD⁡(Xˉ)=σn.E[\bar X]=\mu,\qquad \operatorname{Var}(\bar X)=\frac{\sigma^2}{n},\qquad \operatorname{SD}(\bar X)=\frac{\sigma}{\sqrt n}.

For 25 independent exam scores with mean 70 and standard deviation 20, the average has mean 70, variance 400/25=16400/25=16, and standard deviation 44. Its coefficient of variation is 4/70≈0.057=5.7%4/70\approx0.057=5.7\%.

Key takeaways

  • Expectations combine linearly; constants shift or scale the mean in the same way as the variable.
  • Variance scales by the square of a multiplier; shifts do not affect variance.
  • For dependent variables, include pairwise covariance terms; independence removes them.
  • Negative correlation can reduce a portfolio’s risk through diversification.
  • Independent averaging preserves the mean while reducing variance by a factor of nn.

Test yourself on this module

Questions from these notes, one at a time, with feedback. Your weak topics are saved.

Set it up free Takes about a minute. Free plans work.