When volatility shifts, so does the probability of catastrophe. Ed Peters and Bruno Miranda of First Quadrant explain why leaning on long-term averages is a habit that can hurt you.
There's a flaw hiding inside most investors' standard risk toolkit. For decades, the go-to move has been to anchor on long-term average standard deviation and treat the odds of a catastrophic market move as more or less fixed over time. Ed Peters and Bruno Miranda think that's wrong. Tail risk isn't a permanent feature of markets. It moves. It clusters. And it does so in ways tied pretty directly to whatever volatility regime the market is currently in.
To track that regime, the authors build what they call the Composite VIX, or CVIX. It's a weighted blend of implied volatility from the S&P 500, EuroStoxx, and WTI oil markets. When the CVIX sits above its long-run median, markets are in high-uncertainty territory. Below it, you're in low uncertainty. The difference between those two states is not subtle. Over their 24-year dataset, MSCI World excess returns averaged negative 3.33% annualized in high-uncertainty periods, with a standard deviation of 18.60%. In low-uncertainty periods, returns averaged positive 10.10%, with standard deviation of just 9.46%.
Same index. Two completely different animals.
The Average Is Lying to You
Here's the core problem. When investors talk about a "three-standard-deviation event," they're almost always referencing the long-run population standard deviation, roughly 15% annualized for large-cap equities. A monthly drop of around 13% clears that bar. The trouble is, that number doesn't actually describe either regime. It's what you get when you average two fundamentally different environments together and call it a baseline.
"Tail risk changes over the market cycle in a fairly dramatic way," Peters and Miranda write. Their charts make the point vividly. The high-uncertainty return distribution has a dramatically enlarged negative tail. The positive tail? Basically normal. Flip to low uncertainty and you get the opposite: thinner tails on both sides, a tall peak near the mean.
The numbers that fall out of this are striking. The probability of a decline of 8.66% or greater, a two-standard-deviation drop using the full-period average, is 8.20% during high-uncertainty periods. During low uncertainty, it's essentially zero. Under a normal distribution assumption, you'd expect 4.50%. So the long-term average isn't a conservative estimate. It overstates risk when things are calm and badly understates it when things get dangerous.
A Better Way to Measure
Peters and Miranda introduce conditional kurtosis and conditional skewness to address this directly.
Kurtosis is a measure of how much a distribution's tails differ from those of a normal bell curve.
Think of it this way. A normal distribution has a kurtosis value of zero. If a dataset's kurtosis is positive, it means the tails are fatter than normal, so extreme outcomes (very large gains or very large losses) happen more often than the bell curve would predict. If kurtosis is negative, the tails are thinner, meaning extreme outcomes are actually rarer than normal.
In the context of the Peters and Miranda paper, high positive kurtosis during high-uncertainty periods (5.17 for the MSCI) is the mathematical confirmation of what investors fear most: when markets are stressed, the probability of a catastrophic move is far higher than standard models suggest. Negative kurtosis during low-uncertainty periods (negative 2.60) means the opposite is true. Calm markets are even calmer than the normal distribution implies, with very large moves in either direction becoming quite rare.
A simple way to picture it: imagine two bell curves sitting side by side. One has a tall, narrow peak and long, heavy tails stretching out on both sides. That's high positive kurtosis. The other has a lower, rounder peak and short, stubby tails. That's negative kurtosis. The first one is the market during a crisis. The second one is the market during a long, quiet bull run.
The idea is straightforward once you see it. Standard statistical packages calculate kurtosis and skewness for a subsample using that subsample's own mean and standard deviation. That masks how different the two regimes actually are from the full-period baseline. The conditional approach anchors both calculations to the mean and standard deviation of the entire time series, then applies them to each regime separately.
The question it answers is precise: "What is the kurtosis given that we are in a high or low uncertainty environment?" In high-uncertainty periods, conditional kurtosis for the MSCI hits 5.17, with statistical significance of 12.92. In low uncertainty, it flips to negative 2.60, meaning tails are actually thinner than normal. Conditional skewness in high uncertainty lands at negative 1.73, highly significant, pointing to real asymmetric downside exposure. In low uncertainty, skewness sits near zero and is statistically insignificant.
And it's not just equities. The pattern shows up across asset classes. REITs and high-yield bonds carry tail risk in high-uncertainty periods that, as the authors put it, "dwarfs that of equities and commodities." Government bonds flip the script. They show positive skewness in high uncertainty, confirming their hedging value, and lose that characteristic when things calm down. Emerging markets and commodities follow the risk-on pattern: heavy left-tail exposure when uncertainty is high, meaningful positive skew when it's low.
Fragile vs. Resilient
Peters and Miranda frame their conclusion around a useful contrast. "The market can be considered to be fragile in high uncertainty where shocks can cause tail events while it is more resilient in the low uncertainty periods." Serial correlation backs this up. Momentum, not mean reversion, drives drawdowns during high-uncertainty stretches. Buying dips is a low-uncertainty playbook. Running it in a high-uncertainty environment is a mismatch.
They're careful not to oversell the framework. The long-run median of the CVIX may shift. The regime definitions used here won't hold forever with precision. But the underlying point stands: "The two regimes are so different from one another that it is unlikely that we are ever at the average, which assumes a level of statistical normality unlikely to exist."
For advisors and allocators who can act on this, the opportunity is real. Adjusting positioning across regimes, not just for volatility but for tail risk, skewness, and correlation simultaneously, is what the authors call conditional strategic asset allocation. It's a meaningful edge for those willing to build it in.
5 Key Takeaways for Advisors and Investors
- Tail risk concentrates in high-uncertainty regimes. The probability of a two-standard-deviation or greater decline is 8.20% when the CVIX is elevated and effectively zero when it's not. The long-term average obscures this almost entirely.
- The VIX is a regime signal, not a volatility forecast. Peters and Miranda are explicit on this point: it's a poor predictor of realized volatility. Its real value is flagging when uncertainty is shifting and tail risk is rising.
- Standard kurtosis and skewness numbers mislead. Calculated the conventional way, they systematically understate tail risk during the periods it's most dangerous. Conditional statistics, anchored to full-period parameters, give a more honest and actionable read.
- Asset classes behave differently across regimes. Government bonds hedge in high uncertainty and lose that quality in calm markets. REITs and high yield carry amplified downside when the CVIX spikes. Knowing the regime changes how you should be positioned.
- Conditional asset allocation is a practical opportunity, not just a theory. Investors who adjust their distributional assumptions by regime, across volatility, tail risk, skew, and correlation, gain an edge that goes well beyond basic risk management.
Footnotes:
1 Peters, Ed, and Bruno Miranda. "How 'Tail Risk' Changes Over the Market Cycle." FQ Perspective, First Quadrant, L.P., June 2026.
