19/07/2024

STATISTICAL MEASURES OF ASSET RETURNS

MEASURES OF CENTRAL TENDENCY AND LOCATION

Measures of central tendency specifies where the data is centred.
Measures of location Quantitative measures that describe the location or distribution of data. They include not
only measures of central tendency but also other measures, such as percentiles.

Measures of central tendency

The arithmetic mean: sum of values of the observations in a dataset divided by the number of observations.
The sample mean: the arithmetic mean value of a sample.

A property and potential draw back of the arithmetic mean is its sensitivity to extreme values/outliers.
Since all observations are equal weighted, the arithmetic mean can be pulled upwards or downwards sharply
because of the presence of outliers.
The median: is the value of the middle item in a dataset, that has been sorted in ascending or descending
order.
In an odd numbered sample of n observations, median is the value of the observation that occupies (n+1)/2
position.
In an even numbered sample, median is the mean of the observation that occupies the position n/2 and (n+2)/2.
A distribution will have only one median.
An advantage compared to mean is that, the outliers don’t affect it.
It doesn’t use all the information about the size of the observations and only focuses on the relative position of
the ranked observations
It is less mathematically tractable than a mean.
The mode: The mode is the most frequently occurring value in a dataset. A dataset can have more than one
mode or no mode at all.
unimodal- when a dataset has a single value that is observed most frequently
bimodal- when a dataset has 2 most frequently occurring values.

Dealing with outliers

After checking for errors in a sample with extreme values there may be a possibility of transforming the
variable (log transformation) or replacing it with another variable which serves the same purpose. Else the
below mentioned steps can be taken:
Option 1: Do nothing; use the data without any adjustment. - use if all values are correct & legitimate
Option 2: Delete all the outliers. - trimmed mean A mean computed after excluding a stated small percentage
of the lowest and highest observations.
Option 3: Replace the outliers with another value.- winsorized mean A mean computed after assigning a stated
percentage of the lowest values equal to one specified low value and a stated percentage of the highest values
equal to one specified high value.

Measures of location
Involves identifying values at or below which specified proprtions of the data lie.
Quantiles: A value at or below which a stated fraction of the data lies. Also referred to as a fractile.
a) Quartiles: divide the distribution into quarters
b) Qunitiles: divides the distribution into fifths
c) Deciles: into tenths
d) Percentiles: into hundredths
Interquartile range (IQR): difference between the third quartile and the first quartile, Q3 - Q1
One way to visualise dispersion data across quartiles is to use a diagram called box and whisker chart.
The whiskers are the lines that run from the box and are bounded by the “fences,” which represent the lowest
and highest values of the distribution.

Quantiles in investment practice

Used in portfolio performance evaluation and in investment strategy development and research
use quantiles to rank performance, such as the performance of assets, indexes, and portfolios.
Performance of investment managers are also measured in percentiles to their peers

Dividing data into quantiles based on a specific objectively quantifiable characteristic, such as sales, market
capitalization, or asset size allows analysts to evaluate the impact of that specific characteristic on a quantity
of interest, such as asset returns, sales, growth, or valuation metrics.

MEASURES OF DISPERSION
Dispersion: is the variability around the central tendency.
Mean addresses the returns and dispersion addresses the risk and uncertainty.
Absolute dispersion: amount of variability present without comparison to any reference point or benchmark.
Common measures of dispersion are:
The range: difference between the maxima and minimum values in a dataset
Range = Maximum value - Minimum value
advantage is the ease of computation, disadvantage is that it cannot tell how the data is distributed.
it is sensitive to outliers that may not be a representative of the distribution.
Mean Absolute Deviations: Since sum of deviations around the mean would always be 0, we can examine the
absoulte deviations around the mean.

it uses all the observations in the sample, but is difficult to manipulate mathematically.
Sample variance and sample standard deviation:
a) Sample Variance: average of the squared deviations around the mean

dividing by n-1 improves the statistical properties of the sample variance.

the -1 factor is called the degrees of freedom in estimating the population variance.
b) Sample Standard Deviation: positive square root of the sample variance
 More easily interpreted than variance.
unlike variance the standard deviation is expressed in the same unit of the data itself.
it is usually presented along with the sample mean when summarizing data.
Downside deviation and coefficient of variation
a) Downside deviation: investors are concerned with values of return (other variables) below some other
level than the mean.
the target downside deviation also called as target semi deviation is a measure of dispersion of
observations below a target.
steps to calculate:
specify the target
find the sum of those squared negative deviations from the target
divide the sum by the total no.of observations in the sample minus 1.
finally, take the square root

20/07/2024
b) Coefficient of variation: at times it becomes difficult to interpret standard deviation in terms of
relative degree of variability of different sets of data, may be because the datasets have markedly
different means or they have different units of measurement. Hence coefficient of variation is used.
Relative dispersion: amount of dispersion relative to a reference value or benchmark. Measured using
coefficient of variation: ratio of standard deviation of a set of observations to their mean.

When the observations are returns CV measures the amount of risk (standard deviation) per unit of
reward. (Mean)
if the sample mean is negative, the statistic becomes useless.
CV is a scale free measure and allows the direct comparison of dispersion of across different datasets.

MEASURES OF SHAPE OF A DISTRIBUTION

One important characteristics of distribution which is of interest to the analyst is the degree of symmetry.
If the return distribution is symmetrical about its mean, equal loss and gain intervals exhibit the same
frequencies.
Characteristics of normal distribution:
Its mean, median & mode are equal
It is completely describes by 2 parameters- its mean and variance (or standard deviation)
Skewness
A distribution that is not symmetrical is termed skewed.
A return distribution with positive skew has frequent small loss and a few extreme gains and vice versa, a
negatively skewed distribution has frequent small gains and a few extreme losses
For continuously positive skewed return distribution, mode < median < mean.
For continuously negative skewed return distribution mean < median < mode.
is computed as the average cubed deviation from the mean, standardized by dividing by the standard deviation
cubed to make the measure free of scale
If a distribution is positively skewed with mean > median, then more than half of the deviations from the mean
are negative and less than half are positive.
Hence, if the skewness is positive then the average magnitude of positive deviations is larger than the
magnitude of negative ones.

Kurtosis 21/07/2024
Greater chance of deviations from mean is perceived as a higher risk.
The statistical measure that indicates the combined weight of the tails of a distribution relative to the rest of
the distribution.
A distribution having fatter tails than normal distributions is called leptokurtic or fat tailed,.
A distribution having thinner tails than normal distributions is called platykurtic or thin tailed.
A distribution having similar weights in tails like normal distribution is called mesokurtic.
A fat-tailed (thin-tailed) distribution tends to generate more frequent (less frequent) extremely large deviations
from the mean
A normal distribution has a kurtosis of 3, fat tailed has above 3.0 and thin tailed has less than 3.0.
Sample excess kurtosis: A sample measure of the degree of a distribution’s kurtosis in excess of the normal
distribution’s kurtosis.

Measure of kurtosis is free scale.

Most of the equity return series have been found tailed. If that is not accounted for in the statistical model
then we may have a situation where we would have under estimated the likelihood of very or very good or
very bad outcomes.
Correlation between two variables
Scatter plot: A useful tool for displaying and understanding potential relationships between two variables
Tight clustering signals a potentially stronger relationship between the variables and vice-versa
Inspect for any linear or non linear relationship in the plot.
Observing data located towards the end of the axis gives an idea about the range
If there is an apparent relationship inspecting the scatter plot would help us identify the outliers which would
in turn help us while preparing the financial model.
Covariance and correlation
correlation: measure of the linear relationship between two random variables
covariance: measure of how two variables move together

the above equation is a sample covariance which is the average value of the product of deviation of observation
on two random variables from their sample means.
covariance is a measure of the joint variability a two random variables
If the random variables vary in the same direction then their covariance is positive. If they vary in the
opposite direction their covariance is negative.
Size of the covariance is difficult to interpret it as it involves squared units units of measure. Hence we use
correlation coefficient.
Sample correlation coefficient: standardized measure of how two variables in a sample more together
Expresses the strength of linear relationships b/w 2 random variables

Properties of correlation
A) correlation ranges from -1 to +1 for 2 random variables.
B) a correlation of zero indicates absence of any linear relationship b/w the variables
C) a positive correlation of +1 indicates perfect linear relationship
D) a negative correlation of -1 indicates perfect inverse linear relationship.

Limitations of correlation analysis

Correlation only measures the linear relationships b/w 2 variables it does not take non-linear relationships into
account.
It may also be an unreliable measure when outliers are present as they are very sensitive to outliers.
Correlation does not imply causation also it may not tell the whole story about the data
Spurious correlation: used to refer to the following:.
A) correlation b/w 2 variables reflecting chance relationships in a dataset
B) correlation induced by a calculation that mixes each of two variables with a third variable.
C) correlation b/w 2 variables arising not from a direct relationship b/w them but from their relation to a third
Variable