Quantitative Methods: Central Tendency, Dispersion, Skewness, and Kurtosis
Central Tendency Measures and Properties
Definition of Central Tendency:
Measures of central tendency identify the center of gravity, center of mass, or average location of a data distribution.
In financial and investment applications, central tendency measures (specifically the arithmetic mean) represent expected periodic returns for asset classes, portfolios, or individual investments.
Arithmetic Mean:
Definition: The sum of all observed values divided by the total number of observations ().
Mathematical Formula:
Key Properties:
Sits at the exact mathematical center of the data, meaning positive deviations and negative deviations from the arithmetic mean sum to zero ().
Highly sensitive to extreme values or outliers. A single severe outlier pulls the mean disproportionately toward its value.
The standard measure of central tendency utilized in mathematical statistics and financial return modeling.
Median:
Definition: The value of the middle-ranked observation when data points are arranged in order of magnitude (ascending or descending).
Positional Calculation: For a dataset with observations, the median's rank position is calculated as:
Implementation Rules:
If is odd, the median is the value of the exact middle observation at position .
If is even, the position yields a midpoint; the median is defined as the arithmetic mean of the two central observation values.
Key Properties:
Highly robust against extreme values and outliers because extreme magnitudes do not alter the order or value of the central position.
Ideal measure of central tendency when evaluating distributions containing severe skewness or fat tails (high kurtosis).
Mode:
Definition: The observation value that occurs with the highest frequency in a data set.
Discrete vs. Continuous Data Applications:
Discrete Data: Consists of finite or countable distinct outcomes (e.g., counting the number of rainy days in a month such as , , , or days). The mode is highly effective for discrete data.
Continuous Data: Consists of measurements along a continuous spectrum (e.g., precise rainfall volumes measured in microliters or asset returns calculated to infinite decimal places). Exact identical observations rarely recur, making individual modes ineffective or nonexistent.
Modal Interval / Modal Range: To apply modal concepts to continuous variables, data distributions are partitioned into ranges or intervals (e.g., return ranges between and or and ). The range containing the greatest number of observations is designated as the modal interval.
Multi-Modality:
Unimodal: A distribution containing a single peak or mode.
Bimodal: A distribution containing two distinct values or intervals tied for the highest frequency.
Multimodal: A distribution with three or more modes.
No Mode: Occurs when every observation in a dataset appears with equal frequency.
Key Properties:
Completely unaffected by extreme outliers.
Provides limited mathematical utility in analytical financial modeling compared to the mean.
Techniques for Handling Outliers and Non-Stationary Data
Definition and Impact of Outliers:
An outlier is an observation located a significant distance away from the main cluster of dataset observations.
Numerical Illustration of Outlier Sensitivity:
Baseline Dataset without Outlier: ().
Arithmetic Mean:
Median: Average of 2nd and 3rd observations ( and ) =
Mode: None.
Augmented Dataset with Outlier: ().
Arithmetic Mean: . The mean experiences a severe shift from to , distorting the representation of central tendency.
Median: Value of the 3rd observation () = . The median shifts minimally from to , demonstrating structural stability.
Methods to Mitigate Outlier Distortion:
Trimmed Mean:
Method: Removes a specified percentage of extreme values from both the lower and upper tails of the ranked dataset prior to calculating the arithmetic mean.
Operational Example: A trimmed mean ignores the lowest and highest of observations, evaluating the arithmetic mean strictly on the remaining central dataset interval.
Winsorized Mean:
Method: Replaces extreme outlier values outside specified percentile cutoffs with the exact values of those bounding percentiles, rather than discarding them.
Operational Example: Given a lower boundary at the percentile (e.g., return) and an upper boundary at the percentile (e.g., return):
Extreme negative returns such as or are recoded upward to .
Extreme positive returns such as are recoded downward to .
The arithmetic mean is then computed across all observations using these adjusted values.
Stationarity vs. Non-Stationarity in Time Series Data:
Time Series Definition: A sequence of statistical data points indexed or plotted in chronological order along a time axis.
Stationarity Assumption:
A time series is stationary if its underlying statistical parameters—specifically its mean, variance, and dispersion structure—remain stable and constant over time.
Essential for statistical modeling, ensuring historical sample statistics accurately estimate true population parameters.
Non-Stationarity Characteristics:
A time series is non-stationary if its mean trends over time or if its dispersion/volatility experiences structural shifts or explosions.
Financial Asset Characteristics: Equity market prices typically exhibit non-stationarity, characterized by upward trending means and asymmetric downside volatility expansion.
Analytical Risks: Non-stationarity invalidates standard statistical modeling assumptions, producing non-persistent patterns and unreliable population estimators.
Data Quantiles, Interquartile Range, and Box-and-Whisker Plots
Quantiles Framework:
Quantiles are positional values that divide a ranked distribution of data into equal fractional parts.
Percentiles: Partition data into equal parts. For example, the percentile represents the value below which of dataset observations lie.
Deciles: Partition data into equal parts ().
Quintiles: Partition data into equal parts (). The quintile corresponds to the decile and the percentile.
Quartiles: Partition data into equal parts ().
First Quartile (): percentile point ( of data lies below).
Second Quartile (): percentile point (equals the median).
Third Quartile (): percentile point ( of data lies below).
Interquartile Range (IQR):
Definition: The numerical distance between the third quartile () and the first quartile ().
Mathematical Formula:
Significance: Represents the spread of the middle of the data observations positioned around the median, eliminating tail extreme effects.
Box-and-Whisker Plots:
Visual Structure:
Central Box: Defined from to , explicitly representing the Interquartile Range (IQR) and the middle of data.
Median Bar: A vertical line drawn inside the box at the value of .
Whiskers: Horizontal lines extending outward from the box boundaries to the minimum (smallest) and maximum (largest) observed dataset values.
Interpretation:
Numerical Illustration: Given and , the Interquartile Range is:
Skewness Assessment: Asymmetrical whisker lengths indicate distribution skewness. A right whisker longer than the left whisker reveals positive skewness, which pulls the arithmetic mean above the median.
Dispersion Measures: Range, MAD, and Median Absolute Deviation
Role of Dispersion in Financial Risk:
Dispersion quantifies how widely spread observations (such as manager returns) are around their measure of central tendency.
In investment management, dispersion serves as the primary metric for risk, capturing the probability and extent of deviating from expected returns, particularly on the downside.
Range:
Definition: The total distance between the absolute maximum and minimum observation values in a dataset.
Mathematical Formula:
Operational Example:
Given a manager's highest return of and lowest return of , the range calculation yields:
Limitations: Simplistic metric that relies exclusively on two extreme observations while ignoring the clustering and distribution of all internal data points.
Mean Absolute Deviation (MAD):
Concept: Overcomes simple deviation cancellation () by dropping positive and negative signs, evaluating the average absolute distance of each observation from the sample mean ().
Mathematical Formula:
Mechanism: The vertical absolute value bars transform negative differences into positive distance values prior to summation.
Median Absolute Deviation:
Concept: Evaluates dispersion relative to the median rather than the mean, offering maximum resistance to outlier distortions.
Step-by-Step Procedure:
Determine the median of the underlying dataset.
Calculate the absolute deviation of each observation from the dataset median: .
Rank the calculated absolute deviations in ascending order.
Identify the median value of those ranked absolute deviations.
Variance and Standard Deviation: Population vs. Sample
Fundamental Principles:
Variance and standard deviation represent the standard risk metrics in corporate finance and investment management.
Variance removes negative signs by squaring deviations from the mean.
Units Disconnect: Squaring percentage returns yields percentage squared (), a unit lacking physical or economic meaning. Taking the square root converts the metric back to original percentage units (), yielding standard deviation.
Population Variance and Standard Deviation:
Population Variance Formula: Where represents population variance, is the true population mean, and is total population size.
Population Standard Deviation Formula:
Sample Variance and Standard Deviation:
Sample Variance Formula: Where represents sample variance, is sample mean, and is sample size.
Correction for Bias ( Denominator):
Dividing by instead of adjusts for the lower dispersion typically observed in sample subsets.
This adjustment slightly increases the resulting calculated variance, making an unbiased estimator of true population variance .
Sample Standard Deviation Formula:
Step-by-Step Comparative Example:
Dataset: ().
Sample Mean Calculation:
Squared Deviations from Mean ():
Sum of Squared Deviations =
Sample Variance ():
Sample Standard Deviation ():
Population Variance () (assuming ):
Population Standard Deviation ():
Sanity Check: The numbers sit at an average standard distance of approximately units away from the central mean value of
Scaling Standard Deviation and Volatility Over Time
Linear Scaling Assumptions for Variance:
Variance scales linearly across time periods under two explicit theoretical assumptions:
Constant periodic variance across time.
Independence of periodic returns over time (no serial correlation).
Multi-Period Variance Equation: For a two-day horizon ():
Square Root of Time Rule for Standard Deviation:
Because standard deviation is the square root of variance, it scales across time horizons according to the square root of time ():
Annualization Example:
Given: Daily standard deviation = , annual trading days
Annual Standard Deviation Calculation:
Real-World Market Violations:
Empirical asset return data frequently displays mean reversion over longer time horizons.
As a result, long-term volatility scales up at a slower rate than predicted by the theoretical rule.
Skewness and Distribution Symmetry
Distribution Moments Context:
Distributional characteristics are categorized into statistical moments:
1st Moment: Mean (Location / Central Tendency)
2nd Moment: Variance (Dispersion / Risk)
3rd Moment: Skewness (Asymmetry)
4th Moment: Kurtosis (Peakedness and Tail Weight)
Definition of Skewness:
Skewness measures distribution asymmetry by calculating the standardized average cubed deviation from the mean.
Cubing mathematical properties preserve algebraic signs while disproportionately magnifying extreme outlier deviations.
Symmetrical Distribution:
Skewness equals zero ().
A symmetric curve (e.g., standard normal distribution) displays equal tail lengths and outlier frequencies on both sides of the mean.
Positive Skewness:
Structure: Features an elongated right-hand tail caused by extreme positive outliers (large gains).
Central Tendency Order (Alphabetical Rule):
Investment Context: Preferred by investors (e.g., venture capital, private equity returns) where upside potential is unbounded while downside loss is limited to .
Negative Skewness:
Structure: Features an elongated left-hand tail caused by extreme negative outliers (large losses).
Central Tendency Order:
Investment Context: Undesirable exposure requiring extra return risk premiums.
Market Occurrence: Common in credit risk portfolios (steady yield punctuated by rare default losses), equity market panics (gradual market rises followed by sudden crashes), and relative value hedge fund strategies.
Kurtosis, Tail Risks, and Excess Kurtosis
Definition of Kurtosis:
Kurtosis measures distribution peakedness and tail fatness based on standardized average deviations raised to the fourth power ().
Power of four transformations amplify tail extremes, serving as a primary metric for tail risk.
Baseline Normal Kurtosis:
A standard normal bell curve possesses an absolute kurtosis value equal to three ().
Excess Kurtosis Formula:
Distribution Classifications by Kurtosis:
Leptokurtic:
Characteristics: High kurtosis () and positive excess kurtosis ().
Shape: Highly peaked center with fat tails.
Risk Profile: Elevated probability of extreme tail events (large losses or gains) compared to a normal distribution.
Platykurtic:
Characteristics: Low kurtosis () and negative excess kurtosis ().
Shape: Flatter, domed peak with thin tails (resembling a duckbill platypus profile).
Mesokurtic:
Characteristics: Kurtosis exactly equals three (), matching a standard normal distribution.
Critical Risk Combination:
Combining negative skewness with high leptokurtosis exposes portfolios to extreme left-tail fat-tail losses.
Empirical Applications and Analytical Problem Sets
S&P 500 Five-Year Return Dispersion Dataset:
Given Parameters: Sample mean , median return = . Five-year annual returns include maximum = and minimum = .
Range Calculation:
Mean Absolute Deviation (MAD) Calculation:
Absolute distances from mean : , ,
Sum of absolute deviations divided by observations yields
Median Absolute Deviation Calculation:
Absolute distances from median : ,
Ranked absolute distances: lowest = , 3rd middle ranked distance = , highest =
Median Absolute Deviation =
Brent Crude Oil Futures Skewness Analysis:
Scenario: 10 years of daily returns for front-month Brent crude oil futures contracts yield a skewness statistic of
Evaluation: Distribution is non-normal, asymmetrical, and exhibits a longer left tail than a normal distribution.
Monthly Return Kurtosis Analysis:
Scenario: 5 years of monthly asset returns yield a kurtosis statistic of
Evaluation: Because , the distribution displays leptokurtosis with positive excess kurtosis (), characterized by fatter tails and more extreme values than a normal distribution.
Private Equity Fund Return Performance (250 Funds):
Scenario: Unimodal distribution of private equity funds with arithmetic mean = and geometric mean =
Evaluation: Positive manager return asymmetry shifts mass into the right tail, resulting in more mass in the right tail than the left tail.
Central Tendency Selection under High Skewness and Kurtosis:
Scenario: Identify the optimal measure of central tendency for a dataset containing severe skewness and high kurtosis.
Evaluation: The median is selected as the most suitable measure because positional ranking insulates it from severe tail outliers and asymmetric skew, whereas the arithmetic mean is heavily distorted.