Reading 7: Statistical Concepts and Market

Returns
 Mean—I
The formulae for the population mean and a sample mean are essentially the same;
both are arithmetic means:

Population Mean: N

X i
= i =1
N

Sample Mean: n

X i
X= i =1
n
 Mean—II
Three other types of mean are the weighted average (weighted mean), geometric mean, and
harmonic mean:
n
Weighted Average (Weighted Mean): XW =  w i X i
i =1

Geometric Mean: 1/ n
 n 
X n = ( X1 X 2 Xn ) =   Xi 
1/ n
G= n X1 X 2
 i =1 
n
Harmonic Mean: XH = n
1
X
i =1 i
 Mean—III
The geometric mean is used to calculate average (compound) returns:

1 + rG = n
(1 + r1 )(1 + r2 ) (1 + rn )
1/ n
 n 
rG =   (1 + rt )  −1
 t =1 
The harmonic mean can be used to compute the average price per share paid when the same value
of shares is purchased at several different per-share prices.

Note that:

X H  XG  X
 Mean—IV
Example: A portfolio has annual returns of +5%, +2%, –4%, and +10%. What is the average
(compound) annual return?

rG = (1 + r1 )(1 + r2 ) (1 + rn ) − 1
n

= 4 (1.05 )(1.02 )( 0.96 )(1.10 ) − 1

= 4 (1.05 )(1.02 )( 0.96 )(1.10 )  = 0.0312 or 3.12%

Note - the arithmetic average of the returns—(5% + 2% + –4% + 10%) / 4 = 3.25%—is greater than the geometric mean.
Example: You make three £10,000 purchases of XYZ stock: one at £25/share, one at
£30/share, and one at £35/share. What average per-share price did you pay?
3
XH = = £29.44
1 +1 +1
£25 £30 £35
Note that the arithmetic mean of the share prices—(£25 + £30 + £35)/3 = £30—is greater than the harmonic mean.
 Practice Question
Annual returns on a portfolio for the last six years have been 12%, 5%, –2%, 10%, –1%, and 1%. The
arithmetic and geometric mean returns, respectively, are closest to:

Arithmetic Geometric
A. 4.167% 3.260%
B. 4.167% 4.031%
C. 5.167% 4.031%
 Practice Question
Annual returns on a portfolio for the last six years have been 12%, 5%, –2%, 10%, –1%, and 1%. The
arithmetic and geometric mean returns, respectively, are closest to:

Arithmetic Geometric
A. 4.167% 3.260%
B. 4.167% 4.031%
C. 5.167% 4.031%

Correct answer: B. 4.167% 4.031%

Arithmetic mean =
(12% + 5% + –2% + 10% + –1% + 1%)/6 = 4.167%

Geometric mean =
[(1.12)(1.05)(0.98)(1.10)(0.99)(1.01)]1/6 – 1 = 4.031%
 Median and Mode
The median is the middle value in a set of data: half the values are greater than the median, half are
less:
• With an odd number of observations, the median is the middle one:
–9%, –9%, –2%, 1%, 1%, 3%, 6%
median = 1%
• With an even number of observations, the median is the arithmetic average of the middle two:
–9%, –6%, –5%, –4%, –2%, 6%, 7%, 9%
median = (–4% + –2%)/2 = –3%
Note that the median is not affected by extreme values (see skewness slide NEAR END), whereas mean values are.

The mode is the value that occurs most often:

–9%, –7%, –4%, –2%, –1%, –1%, 0%, 6%
mode = –1%
Data sets can have more than one mode (bimodal, etc.).
 Quantiles (Fractiles)
Percentiles are relatively common. It’s a 2-step process: Step 1 (location); Step 2 (value)

For a data set with n values, the position (location) of the yth percentile is calculated as:
y
Ly = ( n + 1)
100
For example, with 57 values, the 28th percentile is located at position (location):
y
Ly = ( n + 1)
28
57 = 16 .24 24% of the way between
100 the 16th and 17th values.

Step 1: The location of yth percentile is just that: a formula for the location (in an ordered list), not a formula for the
value.

Step 2: If the 16th value were 10 and the 17th value 12, then the 28th percentile would have the location of 16.24 and the
value of 10.48 (which is 24% of the way from 10 to 12); .48 = Delta of 2 (12 – 10) times .24
 Even number of
Practice Question observations
Given 12 annual returns of –3%, –2%, 1%, 1%, 4%, 6%, 8%, 8%, 9%, 9%, 11%, and 15%, the median
return and the value of the 35th percentile are, respectively, closest to:

Median 35th Percentile

A. 7% 1.60%
B. 7% 2.65%
C. 9% 2.65%
 Practice Question
Given 12 annual returns of –3%, –2%, 1%, 1%, 4%, 6%, 8%, 8%, 9%, 9%, 11%, and 15%, the median
return and the value of the 35th percentile are, respectively, closest to:

Median 35th Percentile

A. 7% 1.60% Note that observations are
B. 7% 2.65% already in order (expect the
C. 9% 2.65% worst on exam day).

Correct answer: B. 7% 2.65%

The 4th observation is 1%.

Median = (6% + 8%)/2 = 7%

35th Percentile location = (12 + 1) × 0.35 = 4.55

35th Percentile value = 1% + 0.55(4% – 1%) = 2.65%
This is the delta.
 Measures of Dispersion—II
Annual returns: –9%, –7%, –6%, –5%, –2%, 5%, 7%, 9%
Note: Mean return = –1%.

Range: The difference between the highest value and the lowest value: 9% – (–9%) = 18%.
Note: Range is easy to use but not very useful, as it ignores all of the measurements other than the two extremes.

Mean Absolute Deviation (MAD): Use absolute values to eliminate the negative signs:
If we don’t use absolute values, then summing the differences of the observations and the mean always gives zero.
n

X i −X Although MAD is a perfectly valid

measure of dispersion, it doesn’t have
MAD = i =1
many of the nice analytical properties that
MAD =
n variance and standard deviation have.

[|–9% – (–1%)| + |–7% – (–1%)| + . . . + |9% – (–1%)|]/8

= 6.125%
 Measures of Dispersion—III
Annual returns: –9%, –7%, –6%, –5%, –2%, 5%, 7%, 9%
Note: Mean return = –1%.

Variance: Use squared values to eliminate the negative signs:

N
Variance of a Population:
 ( Xi −  )
2

2 = i =1
N

Variance of a Sample:
( X )
n 2
i −X
s2 = i =1
n −1
 Measures of Dispersion—IV
Annual returns: –9%, –7%, –6%, –5%, –2%, 5%, 7%, 9%
Note: Mean return = –1%.

Example: If the annual returns given are the entire population:

( −9% − ( −1%))2 + ( −7% − ( −1%))2 + + (9% − ( −1%))2

 =
2

8
= 0.004275

If the annual returns given are a sample:

( −9% − ( −1%))2 + ( −7% − ( −1%))2 + + (9% − ( −1%))2

s =
2

7
= 0.004886
 Measures of Dispersion—V
Annual returns: 9%, –7%, –6%, –5%, –2%, 5%, 7%, 9%
Note: Mean return = –1%.

The units of variance are the square of the units of the underlying measurements. To overcome
that incompatibility, we have:

Standard Deviation: the square root of variance

Standard deviation of a population:  = 2

Standard deviation of a sample: s = s2

Standard deviation has the same units as the original data.

 Measures of Dispersion—VI
Annual returns: –9%, –7%, –6%, –5%, –2%, 5%, 7%, 9%
Note: Mean return = –1%.

Example: If the annual returns given are the entire population:

 =  2 = 0.004275 = 6.54%

If the annual returns given are a sample:

s = s 2 = 0.004886 = 6.99%
 Practice Question
Given a sample of 4 annual returns of –3%, 1%, 4%, and 6%, the MAD and standard deviation are,
respectively, closest to:

MAD Standard Deviation

A. 3.00% 3.39%
B. 3.00% 3.92%
C. 4.00% 3.92%
 Practice Question
Given a sample of 4 annual returns of –3%, 1%, 4%, and 6%, the MAD and standard deviation are, respectively,
closest to:

MAD Standard Deviation

A. 3.00% 3.39%
B. 3.00% 3.92%
C. 4.00% 3.92%

Correct answer: B. 3.00% 3.92%

Mean = (–3% + 1% + 4% + 6%)/4 = 2%

MAD = (|–3% – 2%| + |1% – 2%| + |4% – 2%| + |6% – 2%|)/4 = 3%

( −3% − 2%) + (1% − 2% ) + ( 4% − 2% ) + ( 6% − 2% )

2 2 2 2

s=
3
= 3.92%
 Coefficient of Variation
The coefficient of variation (CV) and the Sharpe ratio are closely related; both are based on the
mean return of a security or portfolio and its standard deviation of returns:
s
Coefficient of Variation =
X
s = security's standard deviation of returns
X = security's mean return
 Coefficient of Variation, Sharpe Ratio—II
CV is a measure of risk per unit of return; a lower CV is better than a higher CV.
Example: Suppose that the risk-free rate is 2%.

Investment B is superior.
 Skewness—I
Normal distributions (and other symmetric distributions) have zero skewness.

Positive skewness indicates

asymmetry with more (or more
extreme) outliers in the right
tail:

Negative skewness indicates

asymmetry with more (or more
extreme) outliers in the left tail:

Absolute skewness > 0.5 is significant.

 Skewness—II
For a positively skewed distribution with a single mode:

mean > median > mode

Recall that the median is not affected by extreme values

For a negatively skewed distribution with a single mode:

mean < median < mode

 Kurtosis—I
Kurtosis measures whether the peak of the distribution is higher than that of a normal distribution
(kurtosis > 3) or less than that of a normal distribution (kurtosis < 3).

A distribution’s kurtosis
minus 3.0 is called its
excess kurtosis.

Absolute excess kurtosis >

1.0 is significant;
i.e., kurtosis > 4.0 or
kurtosis < 2.0.
 Kurtosis —II
A distribution with positive excess kurtosis (kurtosis > 3) is called leptokurtic; it also
tends to have fatter tails than a normal distribution.

A distribution with negative excess kurtosis (kurtosis < 3) is called platykurtic; it also
tends to have thinner tails than a normal distribution.

A distribution with zero excess kurtosis (kurtosis = 3) is called mesokurtic.

 Practice Question
Investments X, Y, and Z all have the following characteristics:

For an investor who is most concerned about not losing principal, the best investment and the
worst investment are likely to be, respectively:

Best Worst
A. X Z
B. Y X
C. X Y
 Practice Question
Investments X, Y, and Z all have the following characteristics:

For an investor who is most concerned about not losing principal, the best investment and the
worst investment are likely to be, respectively:

Best Worst Correct answer: A. X Z

A. X Z X has positive skewness and negative excess kurtosis, both of
B. Y X which lead to less probability of (extreme) negative returns. Z
C. X Y has essentially zero skewness, but its positive excess kurtosis
leads to a higher probability of extreme negative returns.