Professional Documents
Culture Documents
19. False. The SD of the total is 2 times σ. The variance of the total sales is 2 times σ 2 .
20. False. The variances are the same. If X1 and X2 are iid, then Var(X1 – X2) = Var(X1 + X2).
21. False. If the effect were simply to introduce dependence (but not otherwise alter the means and SDs), then the
expected difference would remain zero.
22. True. The variance of a difference (which is a weighted sum) depends on the covariance.
Think About It
23. Negative covariance produces smaller variance for the portfolio, and hence less risk.
24. Not necessarily. The variance depends on the covariance among the investments.
25. The covariance between Y and itself is its variance. The covariance between X and Y is the expected value
E(X−µx)(Y − µy). If the two random variables are the same, then E(Y−µy)(Y − µy) = E(Y − µy)2 = Var(Y). The
correlation is 1.
71
26. No. The Sharpe ratio is useful for comparing portfolios with different investments, not for deciding how much
to invest in any one. These two alternatives have the same Sharpe ratio, just as we found that S(X) = S(2X) in
this chapter.
27. No. The covariance depends on the units used to measure the investments. If we track the prices in pennies, say,
rather than dollars, we can make the covariance get very large.
28. Yes. You could predict the price of one from the other with perfect accuracy. The joint distribution of the two
would be concentrated along the positive diagonal.
29. Not likely. Sales during the weekend would probably look rather different than those during the week.
30. In either case, we should predict 0 for the change tomorrow, the expected value. Knowing the change today has
no effect on the distribution tomorrow if the changes are independent.
31. (a) Positively correlated. It seems unlikely that a homeowner would spend a lot for labor but scrimp for the
appliances. An elaborate project would generate higher costs in both categories, unless there are budget
constraints.
(b) A budget constraint may produce negative association or independence. If the family spends a lot on
appliances, for example, then they will not have much left over for labor.
32. The positive dependence produces more variation in the total amounts paid for accidents. Some accidents have
very low payouts with little damage to property or person, whereas others have high payouts in both. Positive
dependence does not affect the expected total paid.
You Do It
33. (a) E(2X – 100) = 2E(X) – 100 = 1,900
Var(2X – 100) = 4Var(X) = 4(200)2 = 160,000; SD(2X – 100) = 400
(b) E(0.5Y) = 0.5E(Y) = 1,000
Var(0.5Y) = 0.25Var(Y) = 0.25(600)2 = 90,000; SD(0.5Y) = 300
(c) E(X + Y) = E(X) + E(Y) = 3,000
Var(X + Y) = Var(X) + Var(Y) = 2002 + 6002=400,000; SD(X + Y) ≈ 632.5
(d) E(X – Y) = E(X) – E(Y) = –1,000
Var(X – Y) = Var(X) + Var(Y) = 2002 + 6002=400,000; SD(X + Y) ≈ 632.5
34. (a) E(8X) = 8E(X) = 800
Var(8X) = 64Var(X) = 64(16)2 = 16,384; SD(8X) = 128
(b) E(3Y – 2) = 3E(Y) – 2 = –152
Var(3Y – 2) = 9Var(Y) = 9(25)2 = 5,625; SD(3Y – 12) = 75
(c) E(X + Y)/2= 0.5E(X) + 0.5E(Y) = 25
Var[(X+Y)/2] = ¼[Var(X)+Var(Y)] =(162 + 252)/4 = 220.25; SD((X + Y)/2) ≈ 14.84
(d) E(X – 2Y) = E(X) – 2 E(Y) = 200
Var(X – 2Y) = Var(X) + 4Var(Y) = 162 + 4(252) = 2,756; SD(X – 2Y) ≈ 52.5
35. All of the calculated expected values remain the same. Only the variances and standard deviations change. Unless both
X and Y appear, the variance is the same.
(a) Unchanged
(b) Unchanged
(c) Var(X + Y) = Var(X) + Var(Y) + 2Cov(X, Y)=2002 + 6002 + 2 × 12,500=425,000; SD(X + Y) ≈ 651.9
(d) Var(X – Y) = Var(X) + Var(Y) – 2Cov(X, Y)=2002 + 6002 – 2 × 12,500=375,000; SD(X + Y) ≈ 612.4
36. All of the calculated expected values remain the same. Only the variances and standard deviations change.
Unless both X and Y appear, the variance is the same. The covariance between X and Y is obtained from ρσxσy =
0.5(16)(25) = 200.
(a) Unchanged
(b) Unchanged
(c) Var((X + Y)/2) = ¼[Var(X) + Var(Y) + 2Cov(X, Y)] = (162 + 252 + 400)/4 = 320.25; SD ≈ 17.90
(d) Var(X – 2Y) = Var(X) + 4Var(Y) – 4Cov(X, Y) = 162 + 4(252) – 800 = 81,956; SD(X – 4Y) ≈ 44.23
37. From the formula for the variance of a sum, the covariance is half of the difference, Cov(X, Y) = ½ [Var(X + Y)
– Var(X) – Var(Y)]= ½ (8 – 10) = −1. The correlation is Cov(X, Y)/[SD(X)SD(Y)] = −1/5 = −0.2.
38. Zero. Treat a constant as a random variable with no variance. Then Cov(X, (c)= E(X−µx)(c – E((c)) = 0.
39. (a) These are dependent, because for example we can write Y = 60 – X. Once we know X, Y is known as well
from this formula.
(b) The covariance lowers the variance. After all, X + Y = 60 and so the variance is 0.
40. (a) These are dependent because the sum of the two must equal 24 hours. For example, we can write Y = 24 – X,
so once we know X, Y is known as well.
(b) Note that X and Y have the same variance, Var(Y) = Var(24 – X) = Var(X). Call this common variance σ .
2
Now work with the random variables directly rather than directly using the covariance. You get the same
answer, but this method is easier. The difference is X – Y = X – (24 – X) = 2X – 24. Hence, the variance of the
difference is 4 σ and the SD is 2σ.
2
41. (a) Let X1 and X2 denote the deliveries for the two. Both drivers are said to operate independently. We assume,
given nothing else, that the number of deliveries is comparable as well.
(b) E(X1 + X2) = 6 + 6 = 12 deliveries
SD(X1 + X2) = 2 2 + 2 2 = 8 ≈ 2.83 deliveries
(c) E(X1 + 1.5X2) = 6 + 1.5 × 6 = 15 hours
(d) SD(X1 + 1.5X2) = 2 2 + (1.52 )(2 2 ) = 13 ≈ 3.61 hours
(e) Yes. It suggests that the counts of the number of deliveries are negatively correlated and not independent, as
if the two drivers split a fixed number of deliveries each day.
42. (a) Let X1 and X2 denote the winnings of the two players. We can assume these are independent since they play
at different sites.
(b) E(X1 + X2) = 300 + 300 = $600
SD(X1 + X2) = 1002 + 1002 = 20,000 ≈ $141.42
(c) E(X1 – X2) = 300 – 300 = $0
SD(X1 – X2) = $141.42 (same as in part b)
(d) First, we may expect the quality of the other players to be better than usual, though this may not be the case.
If so, then the mean winnings for each may fall and the variance increase. Second, by playing at the same table,
for one to win means that the other must lose. This would lead to a negative correlation between X1 and X2.
43. (a) E(X) = 1(0.50) + 2(0.50) = 1.5 sandwiches
Var(X) = (1 − 1.5)2(0.50) + (2 − 1.5)2(0.5) = 0.25 sandwiches2
(b) E(Y) = 1(0.60) + 2(0.35) + 3(0.05) = 1.45 drinks
Var(Y) = (1 − 1.45)2 (0.60) + (2 − 1.45)2 (0.35) + (3 − 1.45)2 (0.05) = 0.3475 drinks2
(c) E(XY) = 1(0.4) + 2(0.2 + 0.1) + 4(0.25) + 6(0.05) = 2.3
Cov(X, Y) = 2.3 – 1.5(1.45) = 0.125 sandwich-drinks
Corr(X, Y) = Cov(X, Y)/(σxσy) = 0.125/(0.5 × 0.589) ≈ 0.424
(d) Customers who buy more drinks also buy more sandwiches. The two are positively related, so those who
buy more of one also purchase more of the other.
(e) E(1.5X + 1Y) = 1.5(1.5) + 1(1.45) = $3.70
Var(1.5X + 1Y) = 1.52Var(X) + Var(Y) + 2(1.5)(1)Cov(X, Y) = 2.25(0.25) + 0.3475 + 3(0.125) = 1.285 $2
Hence, SD(1.5X + 1Y) ≈ $1.13.
(f) E(Y/X) = 1(0.40 + 0.25) + 2(0.1) + 0.5(0.2) + 1.5(0.05) = 1.025
The ratio of means, µy/µx = 1.45/1.5 ≈ 0.97, is less than 1. These do not agree. In general, for positive random
variables, E(Y/X) ≥ E(Y)/E(X).
44. (a) E(X) = P(X = 1)= 0.65 policies
Var(X) = (0 − 0.65)2 (0.35) + (1 − 0.65)2 (0.65) = 0.2275 policies2
(b) E(Y) = P(Y = 1) = 0.65 policies
Var(Y) = (0 − 0.65)2 (0.35) + (1 − 0.65)2 (0.65) = 0.2275 policies2
(c) E(XY) = 0.4 = P(X = 1 and Y = 1) = p(1, 1), the probability that both X and Y are 1. All of the other terms
drop out of the sum.
(d) Cov(X, Y) = E(XY) – µxµy = 0.4 – (0.65)(0.65) = –0.0225. Notice that E(XY) – µx µy = p(1, 1) –px(1) py(1) =
0 if X and Y are independent. These nearly are. The correlation is also small:
Corr(X, Y) = Cov(X, Y)/(σxσy) ≈ –0.0225/(0.2275) = –0.0989.
(e) The small but negative correlation indicates that the agent is not doing so well selling both types of policies.
The sales of the policies are almost independent.
(f) E(750X + 300Y) = 750E(X) + 300E(Y) = 750(0.65) + 300(0.65) = $682.5
Var(750X + 300Y) = 7502Var(X) + 3002Var(Y) + 2(750)(300) Cov(X, Y)
= 7502(0.2275)+ 3002(0.2275) − 2(750)(300)(-0.0225) = 158,568.75$2
SD(750X + 300Y) ≈ $398
45. (a) From the numbers given for the 2004 to 2005 season, we get
E(X) = 0(689/1376) + 2(687/1376) ≈ 0.999
Var(X) = (0 − 0.999)2 (689/1376) + (2 − 0.999)2(687/1376) ≈ 1.000
(b) E(Y) = 0(200/308) + 3(108/308) ≈ 1.052
Var(Y) = (0 − 1.052)2(200/308) + (3 – 1.052)2(108/308) ≈ 2.049
(c) Let X1, X2 … X20 denote the 20 attempted two-point shots and let Y1, Y2 , … ,Y5 denote the three-point
attempts. Then the total number of points scored is
T = X1 + X2 + … + X20 + Y1 + … + Y5
Assuming the X’s are identically distributed and the Y’s are identically distributed,
E(T) = 20µx + 5µy ≈ 20(0.999) + 5(1.052) = 25.24
(d) To compare, we need the standard deviation of the points scored from 2 and 3-point baskets. If we assume
independence of the shots, then
Var(T) = 20Var(X1) + 5Var(Y1) ≈ 20(1.000) + 5(2.049) = 30.245
where we get the variances of these from the calculations in parts a and b.
Hence, we can see that the SD(T) ≈ 5.5. His scoring in this game is within 1 SD of his season performance, so it
seems typical from what we are given.
46. (a) E(X) = 0(1/9) + 3(8/9) ≈ 2.67 points
Var(X) ≈ (0 – 2.67)2(1/9) + (3 − 2.67)2(8/9) = 0.8889
(b) E(Y) = 0(4/15) + 3 (11/15) = 2.2 points
Var(Y) = (0 – 2.2)2(4/15) + (3 − 2.2)2(11/15) ≈ 1.760
(c) Let X1 , X2, and X3 denote the three shorter kicks, and Y1 and Y2 denote the longer kicks. Then the total
number of points scored (call it T) is T = X1 + X2 + X3 +Y1 + Y2 and E(T) = 3(2.67) + 2(2.2) = 12.41 points.
(d) We need the standard deviation of T to see if this score is close to his usual total of 6.6. If we assume the
outcomes are independent, then
Var(T) = 3Var(X1) + 2Var(Y1) ≈ 3(0.8889) + 2(1.760) ≈ 6.1867
and SD(T) ≈ 2.487 points. The expected total in this game is (12.41 − 6.6)/2.487 ≈ 2.34 standard deviations
higher than his typical total for the season. That’s a high score for him.
47. (a) No. Larger homes with more occupants would typically use more of both. Expect positive dependence.
(b) The total cost is T = 0.09X + 10Y . Hence the expected total cost is
E(0.09X + 10Y) = 0.09(12,000) + 10(85) = $1,930
(c) The covariance between X and Y is ρσxσy = 0.35(2,000)(15) = 10,500 KWH−MCF
Using this, we can find that
Var(T) = 0.092Var(X) + 102Var(Y) + 2(0.09)(10)Cov(X, Y)
= 0.092(2,000)2 + 102(15)2 + 2(0.09)(10)10,500 = 73,800 $2
so that SD(T) ≈ $272.
(d) The costs for these homes are only about (1,930 − 1,640)/272 = 1.1 SDs above the national figure for 2001.
That was a long time ago, so these numbers seem atypical.
48. Let X denote sales under the new program and Y the standard.
(a) E(X − Y) = 52,000 − 45,000 = $7,000.
If X and Y are independent, then
Var(X – Y) = Var(X) + Var(Y) = 2(6,000)2 = 72,000,000 $2 and SD(X – Y) ≈ $8,485
The SD is larger than the gap between the expected levels of sales, so there is a good chance it would seem,
from what we are given, that the sales representative from the standard program might sell more.
(b) If ρ = 0.8, then the expected difference remains $7,000 but the variance of the difference becomes
4M Real Money
(a) By splitting the investment, the investor obtains a more steady flow of returns (i.e., less variance than if the
money is placed all into one or the other.
(b) The Sharpe ratio is useful because it combines the mean and variance of the investment. It takes account not only
of the growth in most stocks, but penalizes this growth for the cost of borrowing (the risk-free rate and for the
variability of the returns. A stock with smaller average return may have a higher Sharpe ratio than a stock with
higher average return if it has smaller variance.
(c) No. The observed correlation in the returns is 0.3044. This seems too large to think of modeling the percentage
changes as independent.
(d) The observed means and variances are
Mean Var
Citigroup X 0.1066 5.0738
Exxon Y 0.0652 2.2310
The Sharpe ratios for Citigroup and Exxon are