Essence of Diversification

In the previous lesson, we explored how return and volatility influence a portfolio’s performance. We saw that a portfolio’s standard deviation depends heavily on the correlation between its assets. In this lesson, we take a closer look at correlation, its effect on portfolio risk, and how it can be measured even when relationships between assets aren’t perfectly linear.

Diversification occurs when investments do not move perfectly in sync. In finance, it's a risk management strategy: instead of putting all your money into a single company, sector, or asset class, you spread it across multiple investments.

It's important to note that diversification isn't about maximizing returns. Sometimes, focusing on a few high-performing assets can outperform a diversified portfolio in the short term. However, over the long run, a diversified portfolio tends to provide more consistent results and lower risk.

The key to diversification is including assets that behave differently under similar market conditions. For instance, when stock prices rise, bond yields often fall. This is an example of negative correlation. While not every investment in a portfolio will be perfectly negatively correlated, the goal is to combine assets that do not move identically.

Think of it like a panel of 10 movie critics. If all critics have identical tastes, their reviews are perfectly correlated – one review could replace the rest. But if their opinions vary, you gain a broader perspective. Similarly, a portfolio benefits from diverse behavior among its assets.

Correlation ranges from -1 to 1:

• A correlation of 1 means no diversification – assets move exactly together.
• Any correlation less than 1 provides some diversification, even if it's very small.
• As correlation decreases, the diversification benefit increases. A correlation of -1 maximizes diversification and represents a perfect hedge.

Next, we'll explore how correlation values translate into measurable changes in portfolio risk.

Let’s examine how negative correlations affect portfolio volatility using a simple example.

Example: Consider an equally weighted portfolio of two assets, A and B:

• Expected returns: $r_A = 5\%$, $r_B = 11\%$
• Standard deviations: $\sigma_A = 3\%$, $\sigma_B = 8\%$
• Correlations: $\rho_{A,B} = 1.0, 0.5, 0, -0.5, -1$

Using the standard deviation formula for portfolios, we can compute how risk changes with correlation:

• $\rho$ = 1 (perfect correlation): Portfolio volatility equals a weighted average of individual asset volatilities – no diversification benefit.
• $\rho$ = 0.5: Volatility decreases slightly compared to perfect correlation.
• $\rho$ = 0 (uncorrelated): Volatility is lower than the average of the two assets, showing that combining uncorrelated assets reduces risk.
• $\rho$ = -0.5: Volatility drops further as the assets move in opposite directions part of the time.
• $\rho$ = -1 (perfect negative correlation): Portfolio achieves minimum variance. This represents a theoretical perfect hedge – one asset gains exactly what the other loses.

In practice, finding two assets with a correlation of -1 is extremely rare. However, even partial negative correlations improve diversification and lower portfolio risk.

1. Lower correlation = more diversification: The smaller the correlation between assets, the more the portfolio benefits from reduced volatility.
2. Negative correlation is ideal but uncommon: Perfect negative correlation offers maximum risk reduction, but small negative correlations still add value.
3. Standard deviation isn't linear: Changes in correlation do not result in proportional changes in volatility; the relationship follows a curve rather than a straight line.

Even a small reduction in correlation can slightly lower portfolio risk. Combining assets with differing returns can increase the expected portfolio return without increasing risk, improving the portfolio’s Sharpe ratio.

We can also look at diversification in a more quantitative way.

Consider a portfolio equally split between assets A and B:

• With perfect correlation ($\rho = 1$), the portfolio’s standard deviation is 5.5%.
• With zero correlation ($\rho = 0$), the standard deviation drops to 4.272%.

By choosing assets that are less than perfectly correlated, we reduce portfolio volatility. The greater the difference in correlation, the larger the benefit: negative correlations provide the maximum reduction in risk.

A simple formula captures this variance reduction relative to the perfectly correlated scenario. If correlation is 1, the reduction is zero with no diversification benefit. As correlation decreases, standard deviation falls, improving the portfolio’s risk-adjusted return (Sharpe ratio). Zero correlation significantly reduces risk, while negative correlations maximize the benefit.

Example:

• Portfolio P contains only asset A: its variance is simply $\sigma_P^2$ = $\sigma_A^2$.
• Portfolio Q adds asset B with weights $w_A$ and $w_B$ (summing to 1). Its variance becomes:

The diversification benefit exists whenever $\sigma_Q^2 < \sigma_P^2$. With some algebra, we can identify the combinations of weights and correlations that produce this benefit.

For portfolios with three or more assets, these relationships become more complex. We can no longer rely on simple formulas, but volatility can still be computed, and adding assets or adjusting weights can be evaluated for their diversification effect. This will be covered in detail in a Portfolio Management course.

Correlation quantifies how two variables move together. There are three widely used measures:

1. Pearson correlation
2. Spearman correlation
3. Kendall correlation

Pearson Correlation

The Pearson correlation coefficient measures the strength of a linear relationship between two variables:. It ranges from -1 to 1:

• 1: perfect positive linear relationship
• 0: no linear relationship
• -1: perfect negative linear relationship

Intuitively, imagine plotting X versus Y and fitting a straight line. Pearson correlation tells you how tightly the points cluster around that line:

• Perfectly aligned points on an upward-sloping line → +1
• Perfectly aligned points on a downward-sloping line → -1
• Scattered points with no trend → 0

Limitations: Pearson correlation only captures linear relationships, assumes normality, requires independence of observations, and is sensitive to outliers.

Spearman Correlation

Sometimes, relationships are monotonic but nonlinear, meaning one variable consistently increases as the other increases, but not in a straight line. Pearson correlation may underestimate this association.

Spearman correlation solves this by using ranks instead of actual values:

1. Assign ranks to each data point in X and Y (largest value = 1, second largest = 2, etc.).
2. Handle ties by averaging ranks.
3. Apply the Pearson formula to the ranks rather than the original values.

The Spearman correlation thus captures monotonic relationships, whether linear or nonlinear, giving a more general measure of association.

• Calculate differences $d_i$ between ranks for each pair.
• Square and sum these differences to compute the correlation.

This method ensures that even strong nonlinear relationships are recognized, which Pearson might miss.

Example:

If the ranks of X perfectly match the ranks of Y, Spearman correlation equals 1. Any deviations in rank reduce the correlation. This makes Spearman particularly useful for financial markets: for instance, to check whether one security tends to rise whenever another rises, regardless of the exact magnitude. Unlike Pearson, it is robust to skewed distributions and outliers.

Limitations: Spearman cannot detect relationships that are non-monotonic. For instance, if Y is a quadratic function of X, neither Pearson nor Spearman will identify a correlation, even if the relationship is perfectly deterministic.

Kendall Correlation

Another nonparametric method is Kendall's tau, which measures the degree of concordance between pairs of observations. Like Spearman, it doesn't assume linearity or normality and is robust to outliers.

How to compute Kendall’s tau:

1. For each pair of observations $(x_i, y_i)$ and $(x_j, y_j)$, determine if they are concordant or discordant:

   • Concordant: if both $x_i > x_j$ and $y_i > y_j$, or both $x_i < x_j$ and $y_i < y_j$.
   • Discordant: if one of the inequalities is reversed.

2. Count the number of concordant pairs (C) and discordant pairs (D).

3. Calculate Kendall's tau:

Where $n$ is the total number of observations.

Kendall's tau ranges from -1 to 1, similar to other correlation measures:

• 1: perfect agreement (all pairs concordant)
• 0: no association
• -1: perfect disagreement (all pairs discordant)

Kendall's tau is often preferred for small sample sizes and when data contain many ties, as it provides a more accurate measure of association in such cases.

Also note that in finance, Spearman and Kendall correlations are particularly useful when asset returns show nonlinear co-movements or when distributions are non-normal, which is often the case in real markets.