Moments, Skewness & Kurtosis: Data Shape & Spread

TAKEAWAYS:

• Moments (especially mean, variance) reveal data’s central tendency and dispersion, while higher-order moments (3rd, 4th) expose asymmetry and peakedness.
• Skewness measures tail asymmetry: positive skew (long right tail) vs. negative skew (long left tail), critical for risk assessment in loans or stock returns.
• Kurtosis compares tail heaviness to normal distribution: leptokurtic (fat tails) vs. platykurtic (thin tails), used in financial modeling (e.g., NEPSE volatility).
• Percentile coefficients (e.g., kurtosis = P₉₀–P₁₀) avoid raw moment calculations for real data with outliers.
• Real-world: eSewa uses skewness to detect fraudulent transaction spikes; Pathao kurtosis identifies driver availability peaks.
• Exam tip: Always interpret moments/skewness/kurtosis in context (e.g., “positive skew → high-end wages dominate”).

---

1. Moments: The Mathematical Fingerprint of Data

Moments are weighted averages of data values, revealing shape and spread. The n-th moment about a point a is:

$${\mu }_n^{\prime }=\frac{1}{N}\sum _{i=1}^N(x_i-a{)}^n$$

Key Moments

• First moment (mean): ${\mu }_1^{\prime }=\bar{x}$ (central tendency).
• Second moment (variance): ${\mu }_2^{\prime }={\sigma }^2+(\bar{x}-a{)}^2$ (spread).
• Third moment (skewness): ${\mu }_3^{\prime }$ (asymmetry).
• Fourth moment (kurtosis): ${\mu }_4^{\prime }$ (tail heaviness).

Why moments? They generalize mean/variance to higher dimensions, capturing non-normal distributions (e.g., income data).

---

Worked Example: Moments About Arbitrary Point 4

Given distribution:

X | 2  3  4  5  6
f | 1  3  7  2  1

Compute first four moments about a = 4.

Step 1: Compute deviations from a

X   | 2  3  4  5  6
X−4 |−2 −1  0  1  2

Step 2: Calculate powers

n=1: (X−4)¹ = [−2, −1, 0, 1, 2]
n=2: (X−4)² = [4, 1, 0, 1, 4]
n=3: (X−4)³ = [−8, −1, 0, 1, 8]
n=4: (X−4)⁴ = [16, 1, 0, 1, 16]

Step 3: Weight by frequency and sum

μ₁' = (1×−2 + 3×−1 + 7×0 + 2×1 + 1×2)/14 = (−2 −3 +0 +2 +2)/14 = **−1/7 ≈ −0.14**
μ₂' = (1×4 + 3×1 + 7×0 + 2×1 + 1×4)/14 = (4 +3 +0 +2 +4)/14 = **13/14 ≈ 0.93**
μ₃' = (1×−8 + 3×−1 + 7×0 + 2×1 + 1×8)/14 = (−8 −3 +0 +2 +8)/14 = **−1/14 ≈ −0.07**
μ₄' = (1×16 + 3×1 + 7×0 + 2×1 + 1×16)/14 = (16 +3 +0 +2 +16)/14 = **37/14 ≈ 2.64**

Interpretation:

• Mean (μ₁'): Data centers at X ≈ 4.14 (since μ₁' = −0.14 about a=4).
• Variance (μ₂'): Spread is 0.93 around a=4.
• Skewness (μ₃'): Negative skew (left tail longer).
• Kurtosis (μ₄'): Higher than normal (fat tails).

---

Moments About Mean vs. Arbitrary Point

Key Insight: Moments about the mean (central moments) are standardized (e.g., skewness = ${\gamma }_1$), while moments about an arbitrary point a are raw values.

---

2. Skewness: The Tail’s Direction

Skewness measures asymmetry in data distribution. The coefficient of skewness (standardized) is:

$$\text{Skewness}=\frac{{\mu }_3}{{\sigma }^3}\text{or}\frac{3(\bar{x}-\text{Median})}{\sigma }$$

Types of Skewness

• Skewness > 0: Right-tailed (e.g., wealth distribution).
• Skewness < 0: Left-tailed (e.g., exam scores with few low performers).
• Skewness = 0: Perfect symmetry (e.g., normal distribution).

---

Worked Example: Skewness from Moments

Given:

• First four central moments: $\mu_1 = 0$, $\mu_2 = 14.75$, $\mu_3 = 39.75$, $\mu_4 = 152.31$.
• Step 1: Compute $\sigma^2 = \mu_2 = 14.75 \Rightarrow \sigma = \sqrt{14.75} \approx 3.84$.
• Step 2: Calculate skewness: $$\text{Skewness}=\frac{{\mu }_3}{{\sigma }^3}=\frac{39.75}{(3.84{)}^3}\approx \frac{39.75}{56.6}\approx \ast \ast 0.70\ast \ast$$
• Interpretation: Positive skew (0.70) → Data has a longer right tail (e.g., NEPSE stock returns where a few high-value trades dominate).

---

3. Kurtosis: The Tail’s Heaviness

Kurtosis compares a distribution’s tail heaviness to a normal distribution. The excess kurtosis is:

$$\text{Kurtosis}=\frac{{\mu }_4}{{\sigma }^4}-3$$

Types of Kurtosis

• Kurtosis > 0: Leptokurtic (fat tails; e.g., stock market volatility).
• Kurtosis = 0: Mesokurtic (normal tails).
• Kurtosis < 0: Platykurtic (thin tails; e.g., uniform distributions).

---

Worked Example: Percentile Kurtosis

Given hourly wage data (from table in exam question):

| Wage Range (Rs) | Frequency |
|-----------------|-----------|
| 23-27           | 12        |
| 28-32           | 18        |
| 33-37           | 25        |
| 38-42           | 15        |
| 43-47           | 10        |
| 48-52           | 8         |

Step 1: Compute P₉₀ and P₁₀.

• Total N = 12 + 18 + 25 + 15 + 10 + 8 = 88.
• P₁₀ = 8.8 → 28-32 (cumulative: 12 → 30th percentile in 28-32).
• P₉₀ = 79.2 → 43-47 (cumulative: 12+18+25+15=70; 79.2 is in 43-47). Step 2: Calculate kurtosis coefficient: $$\text{Kurtosis}=\frac{P_{90}-P_{10}}{2\times \text{IQR}}=\frac{45-29.5}{2\times (37-23)}=\frac{15.5}{28}\approx \ast \ast 0.55\ast \ast$$ Interpretation: Leptokurtic (0.55) → Wage distribution has fat tails (few earn very high/low wages).

---

4. Real-World Applications

## In the Real World

1. NEPSE Stock Market (Nepal Stock Exchange)

   • Idea: Skewness and kurtosis measure risk.
   • How: Positive skew (right tail) indicates potential for high returns but also higher crash risk (e.g., 2021 NEPSE volatility).
   • Example: If NEPSE’s stock returns have skewness = 0.8, traders expect few extreme gains but prepare for sharp drops.

2. Pathao Driver Availability

   • Idea: Kurtosis models driver supply spikes.
   • How: Leptokurtic kurtosis (fat tails) shows peak-hour surges (e.g., 6–9 PM) and low availability at off-peak times.
   • Worked Example: Suppose Pathao’s driver count per hour follows a distribution with kurtosis = 1.2. This suggests more variability than normal, meaning:
     • 6 PM: High demand → surge pricing.
     • 3 AM: Low demand → fewer drivers online.

3. eSewa Transaction Volumes

   • Idea: Skewness detects fraud patterns.
   • How: Negative skew (left tail) may indicate many small transactions but few large fraudulent transfers.
   • Example: If eSewa’s daily transaction amounts have skewness = −0.5, analysts flag unusual low-value spikes as potential bot activity.

---

5. Comparison Table: Skewness vs. Kurtosis

---

6. Exam Tip: How to Score Full Marks

1. Definitions First:

   • Always define skewness and kurtosis with formulas (e.g., “Skewness = $\frac{3(\bar{x}-\text{Median})}{\sigma }$”).
   • For moments, state: “The n-th moment about a is ${\mu }_n^{\prime }=\frac{1}{N}\sum (x_i-a{)}^n$.”

2. Show Work for Moments:

   • For arbitrary point moments, compute deviations step-by-step (as in the worked example above).
   • For central moments, derive $\sigma$ first, then compute skewness/kurtosis.

3. Interpret Graphically:

   • Draw a skewed curve (left/right) and label tails.
   • For kurtosis, sketch fat vs. thin tails and compare to normal distribution.

4. Contextualize Answers:

   • Link to real data: “Positive skewness in NEPSE returns suggests high-risk, high-reward investments.”
   • For kurtosis: “Leptokurtic data implies unexpected extreme events (e.g., Pathao’s 6 PM rush).”

5. Percentile Kurtosis Shortcut:

   • If given raw data, use P₉₀ − P₁₀ over IQR for kurtosis (as in the wage example).
   • Always round to 2 decimal places and interpret (e.g., “kurtosis = 0.55 → fat-tailed distribution”).

6. Common Pitfalls to Avoid:

   • Mixing moments: Use central moments for skewness/kurtosis; raw moments for arbitrary points.
   • Ignoring units: If data is in Rs, state skewness as “dimensionless” (since it’s standardized).
   • Overlooking assumptions: Skewness/kurtosis assume no outliers (check with boxplots).

---

Sample Exam Answer Structure

Question: Compute skewness and kurtosis for data with moments: ${\mu }_1=0$, ${\mu }_2=16$, ${\mu }_3=48$, ${\mu }_4=200$. Interpret.

Answer:

1. Compute $\sigma$: $$\sigma =\sqrt{{\mu }_2}=\sqrt{16}=4$$

2. Skewness: $$\text{Skewness}=\frac{{\mu }_3}{{\sigma }^3}=\frac{48}{64}=0.75$$ Interpretation: Positive skewness (0.75) indicates a right-skewed distribution (e.g., income data where most earn moderately, but a few earn extremely high salaries).

3. Kurtosis: $$\text{Kurtosis}=\frac{{\mu }_4}{{\sigma }^4}-3=\frac{200}{256}-3=0.78-3=-2.22$$ Interpretation: Negative kurtosis (−2.22) suggests a platykurtic distribution (thinner tails than normal), meaning fewer extreme values (e.g., uniform driver availability in Pathao).

---

Visual Summary

Moments Hierarchy:

┌───────────────────────────────────────┐
│            Moments                    │
├─────────────┬─────────────────────────┤
│ First       │ Second (Variance)       │
│ (Mean)      │ (Spread)                │
├─────────────┼─────────────────────────┤
│ Third       │ Fourth (Kurtosis)       │
│ (Skewness)  │ (Tail Heaviness)        │
└─────────────┴─────────────────────────┘