Welcome to the Real World of Correlation!
In FRM Part I, you learned how to calculate correlation as a neat, clean number between -1 and +1. But in the real world of Market Risk Measurement and Management, correlation is anything but neat. It is "moody"—it changes when the market gets stressed, it moves over time, and it rarely stays where you expect it to be.
In this chapter, we are going to explore how correlations actually behave in financial markets. Understanding these properties is crucial because if your risk models assume correlation is constant, you might be in for a nasty surprise when a crisis hits. Let’s dive in!
1. Correlation is Not Constant (Time-Varying)
The first thing to realize is that correlation is dynamic. If you measure the correlation between two stocks today, it will likely be different from the correlation measured six months from now.
Mean Reversion
While correlation moves around, it generally exhibits mean reversion. This means that if the correlation between two assets becomes unusually high or low compared to its historical average, it tends to pull back toward that average over time.
Analogy: Think of a rubber band. You can stretch it (move correlation away from the average), but eventually, it wants to snap back to its original shape (the long-run mean).
Quick Review: Why does this matter?
If you are a risk manager using a "static" (unchanging) correlation in your Value-at-Risk (VaR) model, you are ignoring the reality that risk can change rapidly. Most modern risk models use GARCH or EWMA (which you saw in Part I) to account for this "moving" nature of correlation.
2. The "Crisis" Effect: Correlation and Volatility
This is perhaps the most famous empirical property of correlation. In the real world, correlations tend to increase during periods of high market volatility and market crashes.
Did you know? In finance, we often say: "The only thing that rises in a falling market is correlation."
Why does this happen?
During a market panic, investors often stop looking at the individual "fundamentals" of a company. Instead, they sell everything to raise cash or exit the market. This "contagion" effect causes almost all risky assets to fall at the same time, driving their correlations toward 1.0.
Key Takeaway: Diversification is most effective when markets are calm. Unfortunately, diversification often disappears right when you need it most (during a crash).
3. Asymmetry in Correlation
Correlations don't just react to volatility; they react to the direction of the market. This is known as asymmetric correlation.
Empirical evidence shows that:
- Correlations between equities are much higher during large market downturns.
- Correlations are generally lower during market upturns (bull markets).
Don't worry if this seems tricky at first! Just remember: Fear is a stronger "unifier" than greed. When people are scared, they all run for the exit at the same time, making stocks move together. When they are happy, they are more selective about which stocks they buy.
4. Tail Dependence: Beyond Linear Correlation
One of the biggest mistakes in risk management is relying solely on Pearson Correlation (the standard \(\rho\) you learned). Standard correlation measures linear relationships, but it fails to capture "tail dependence."
What is Tail Dependence?
Tail dependence is the probability that two assets will experience extreme moves (the "tails" of the distribution) at the same time. Even if two assets have a low average correlation, they might have high lower-tail dependence, meaning they crash together during a "black swan" event.
Common Mistake to Avoid: Assuming that a correlation of 0.30 means you are safe. If the assets have high tail dependence, that 0.30 could jump to 0.90 in a crisis, leading to much larger losses than your model predicted.
5. Properties Across Asset Classes
Correlations behave differently depending on what you are trading. Here is a quick breakdown:
Equities (Stocks)
- Usually positive.
- Very sensitive to market crashes (asymmetric).
- International stock correlations have increased over time due to globalization.
Bonds
- Correlations between high-quality government bonds (like US Treasuries) and stocks are often negative during crises. This is the "flight to quality" effect.
- However, "Junk Bonds" (high-yield) behave more like stocks and see their correlations rise during crashes.
Currencies
- Currency correlations are highly influenced by monetary policy and interest rate differentials.
- They can be very unstable and prone to sudden shifts based on central bank interventions.
Summary Box:
Bull Market: Low Correlation + High Diversification.
Bear Market: High Correlation + Low Diversification.
6. The Impact of Time Horizon
Does it matter if you measure correlation using daily data versus monthly data? Yes!
Empirically, as the time horizon increases, measured correlations tend to increase. This is often because short-term "noise" (random price movements) cancels out over longer periods, revealing the underlying economic links between assets.
Mnemonic: Short-term is noisy, Long-term is cozy. (Correlations look higher/stronger over longer periods because the random daily zig-zags smooth out).
7. Pitfalls in Measuring Correlation
When you are looking at correlation data, be careful of these two traps:
1. The Volatility Bias
Mathematically, if the volatility of a sample increases, the calculated correlation coefficient will often appear higher, even if the underlying relationship hasn't changed. This is a statistical quirk that can lead risk managers to overstate how much the "true" correlation has shifted.
2. Non-synchronous Trading
If you are measuring the correlation between a stock in New York and a stock in Tokyo, the markets aren't open at the same time. This "lag" can make correlations appear lower than they actually are. This is known as the Epps Effect.
Final Quick Review
To wrap up, here are the "Must-Know" points for your exam:
- Mean Reversion: Correlation moves but eventually returns to its average.
- Asymmetry: Correlation is higher in down markets than in up markets.
- Diversification Breakdown: In a crisis, correlations rise toward 1.0, making portfolios riskier than expected.
- Tail Dependence: Linear correlation (\(\rho\)) doesn't catch extreme joint events; we need to look at the "tails."
- Horizon: Longer measurement periods usually result in higher correlation estimates.
Keep going! You've just mastered one of the most practical chapters in Market Risk. Understanding that "numbers change when things get bad" is the hallmark of a great Risk Manager!