Every now and then the market reminds you that the most dangerous risks aren’t the ones you can see coming. They’re the ones embedded in the risk management systems themselves. Value-at-Risk (VaR) shocks are one of those paradoxes in finance where the tools designed to keep portfolios safe can end up making everything worse. Understanding how they work, and why they matter, is essential for anyone managing leveraged portfolios or trying to make sense of the violent dislocations that periodically rip through markets.

The basic idea is straightforward enough. Many portfolios size their positions based on some measure of expected risk – typically volatility, asset correlation, and the resulting VaR. In calm markets, volatility is low, correlations are well-behaved, and risk models give you the green light to run bigger positions. Life is good, until something upsets the apple cart. Volatility spikes, correlations jump, everybody’s risk model says the same thing at the same time – cut risk.

Below is a deliberately simplified illustrative simulation to walk through the mechanics of a VaR shock. Think of it as a stylized version of what happens in the real world when a stress event hits a volatility-targeted portfolio, whether it’s a risk parity fund, a sector and factor neutral fund, a spread trader or a trend follower. Any leveraged strategy that dynamically adjusts exposure based on realized risk metrics.

We use two assets, each running around 15% annualized volatility in normal times, with correlations near zero. The portfolio targets a 15% volatility max and sizes positions accordingly. With two uncorrelated assets at 15% vol each, the math is generous as the diversification benefit from near-zero correlation means you can hold around 70-80% in each asset (roughly 1.5x total exposure) and still run portfolio-level volatility comfortably below the 15% limit, closer to 10%. That’s the beauty of diversification working as intended. Low correlations give you leverage capacity for free. The risk model sees calm vol, calm correlations, and allocates accordingly. Everything is in equilibrium. Now let’s break it.

Volatility Spikes and Correlations Jump

In a stress regime hits (marked by the red dashed line), we increase asset volatility markedly. The models that told you 15% vol was the steady state suddenly need to absorb a 3x move in that input. Seems an extreme jump, but markets do this every now and again. Silver is doing it right now, as are many individual stocks. These kinds of moves are rare but not at all unprecedented, and they are exactly the conditions that stress the assumptions underpinning most risk models.

If we stress the other term at the same time, it gets worse. In the calm regime, the correlation between Asset A and Asset B bounces around near zero – sometimes slightly positive, sometimes slightly negative, averaging out to very little co-movement. This is what diversification looks like on paper. Two assets, similar volatility, low correlation, textbook portfolio construction. This low correlation that allowed the portfolio to run 1.5x total exposure while keeping portfolio vol around 10%, well below the 15% max.

In our stress event we push the asset correlation toward 0.7. In crisis, everything becomes the same trade. We’ve seen this repeatedly: in 2008, what had been nicely diversified positions in equities, FX, commodities, and emerging markets all started moving together. In the Yen carry unwind of August 2024, seemingly unrelated positions across global equity markets unwound simultaneously. Factor portfolios in stocks show similar dynamics. The diversification benefit being relied on can evaporate at the worst time.

This is the double whammy that makes VaR shocks so vicious. It’s not just that individual asset vols are higher. The portfolio-level risk is getting hit from both directions – higher component volatilities and higher correlations between them. The two inputs that gave you all that free leverage capacity just turned against you simultaneously.

Unconstrained Portfolio Vol Explodes

Now let’s see what those two effects – spiking vols and spiking correlations – do to the portfolio if you don’t adjust anything. The blue line shows what the portfolio’s volatility would be if you held your positions constant through the stress event. Two uncorrelated low volatility assets turn into one high volatility asset. The green dashed line is our 15% volatility max. In normal times the unconstrained portfolio vol runs around 10%. When the stress regime hits, portfolio vol jumps quick to almost three times the target risk. For a leveraged portfolio with margin constraints, risk limits, or investor mandates, this is a big problem. Phrase that rings true here…

“It ain’t what you don’t know that gets you into trouble. It’s what you know for sure that just ain’t so.” – Mark Twain

Forced Deleveraging

The risk model forces position cuts across both assets. Before the stress event, position sizes fluctuated around 60-80% in each asset. Once stress hits, both positions get cut back. Asset B, which saw the larger vol spike, gets cut more aggressively – dropping from around 0.8 to under 0.2. Asset A follows a similar trajectory. The portfolio is being forced to sell.

This is the critical mechanism. It’s not a discretionary decision to reduce risk, rather a mechanical response from the risk management system. Every approach running a similar framework sees the same inputs (higher vol, higher correlation) and reacts accordingly to cut exposure. Regular entry and exit investment decisions have different drivers, risk system exits tend to cluster as all try to get through the exit door at the same time. Forced selling pushes prices further, which raises volatility further, which triggers more selling and margin calls.

Total Exposure Drops

Zooming out to total portfolio exposure (the sum of both positions), show the normal regime, total exposure ran between 1.2x and 2.0x – typical for a leveraged vol-targeted strategy. Post-stress, total exposure collapses to around 0.3x to 0.5x. The portfolio has gone from being meaningfully invested to having a large cash position. The strategy that was designed to maintain a consistent risk profile through time ends up with a wildly inconsistent capital deployment profile. You run maximum exposure during calm periods, which in a Minsky model become reinforcing, and then quickly reduce exposure during stress periods (when expected returns are often highest, and assets are on sale). It’s buying high and selling low, executed systematically by a risk model that is doing exactly what it was told to do.

When enough capital is managed with these kinds of vol-sensitive frameworks – and a lot of capital is – the deleveraging becomes a source of selling that has nothing to do with fundamentals. Prices aren’t falling because earnings forecasts changed or because the economic outlook deteriorated. They’re falling because risk models are forcing selling.

Realized Vol Stays Under Control (Mission Accomplished?)

The final chart shows the other side of the coin. After all that deleveraging, the portfolio’s realized volatility is kept near the 15% target throughout the stress period. The risk model did its job.

But at what cost? The portfolio achieved risk control by moving to much reduced exposures at exactly the point when forward-looking returns may be most attractive. It sold the lows. And it will slowly re-lever into the recovery, buying back exposure as volatility declines – which means it buys the highs too. The risk target was met, but the return outcome was degraded by the mechanical response.

This is the fundamental tension at the heart of vol-targeting. The risk management works in a narrow, mechanical sense, but the costs are borne not just by the individual portfolio but by the market ecosystem as a whole, as the forced selling creates cascading effects across correlated assets. What is individually rational becomes collectively suboptimal.

Are these happening more?

It doesn’t seem so, this dynamic has been at play for a long time. The Lowenstein book When Genius Failed is a fantastic blow by blow account of the LTCM episode in 1998. There is always a proximate cause. The Minsky point is that as markets get more complacent, more stretched and show lower volatility the trigger gets closer.

The most recent episode in software and to a lesser extent healthcare seems somewhat driven by a couple of broader macro-level shifts. While the index remains calm currently, under the surface there is a lot of volatility in a few sectors. AI capabilities are rapidly advancing leading to existential fears around terminal values and long-term earnings power in what were previously viewed as high quality, high growth and high margin franchises. Big shifts like this causes volatility and correlation regime shifts and a shoot first ask questions later mentality.

One shouldn’t panic, but if you are going to panic in this VaR shock world, you better be the first one out the door.