Research insight · Issue 010
Markets Share Volatility Differently Under Stress
Cross-market volatility coupling rises around major stress events, but not equally across relationships or horizons.
- Published
- Data through
- 24 September 2026
ContentsExecutive summary
Executive summary
Cross-market volatility coupling rises around major stress events, but not equally across relationships or horizons.
- 01SPY-QQQ is already highly coupled in normal conditions.
- 02QQQ-EEM and SPY-EEM show the largest stress-driven increases.
- 03The effect is strongest over 20-30 trading days.
- SPY-QQQ coupling
- 0.72
- SPY-EEM coupling
- 0.59
- QQQ-EEM stress increase
- +0.22
- SPY-EEM stress increase
- +0.21
- Strongest horizon
- 20-30 days
Why it matters
Some volatility coupling is structural. Other relationships tighten only when stress arrives. That means weaker cross-market relationships can converge quickly when markets come under pressure.
Analysis 01
Some volatility coupling is already structural
SPY-QQQ has the highest average variance-shock coupling at 0.72, ahead of SPY-FEZ (0.60), SPY-EEM (0.59) and QQQ-EEM (0.58). Some relationships are therefore more tightly coupled even before stress arrives. That matters because it shows that volatility coupling is not purely a crisis effect. Some of it is already structural. Stress may intensify these links, but it does not begin from the same baseline in every market pair.
SPY-QQQ has mean coupling 0.72, SPY-FEZ 0.60, SPY-EEM 0.59 and QQQ-EEM 0.58.
Analysis 02
Coupling changes through time
SPY-QQQ remains comparatively elevated across most of the sample, while the international pairs move through larger shifts. The gap therefore changes through time rather than remaining fixed. Cross-market coupling can strengthen materially in some periods and weaken again in others. The relationships are persistent, but their intensity is state-dependent.
SPY-QQQ remains comparatively elevated across most years; international equity pairs show larger shifts in coupling.
Analysis 03
Stress pulls volatility shocks together
Cross-market volatility coupling rises sharply around major stress events. All four pairs move higher as the VIX approaches its peak, then separate again as the episode passes. The response is not identical across relationships. SPY-QQQ remains the most tightly coupled throughout, but the cross-market pairs also converge materially around the stress peak. Stress does not affect volatility in isolation. It increases the extent to which volatility shocks become shared.
All four pairs rise around the VIX peak and subsequently separate. SPY-QQQ generally remains more tightly coupled than the international equity pairs.
Analysis 04
Stress adds the most where coupling starts lower
The largest average increases occur in the cross-market pairs. QQQ-EEM rises by 0.22 and SPY-EEM by 0.21, compared with only 0.08 for SPY-QQQ. The confidence intervals are wider for some pairs, but the contrast is clear. Stress has less incremental effect where coupling is already structurally high. The key distinction is between how volatile markets are and whether their volatility shocks are becoming shared.
QQQ-EEM has a mean increase of about 0.22, SPY-EEM 0.21, SPY-FEZ 0.17 and SPY-QQQ 0.08. The plotted intervals for SPY-FEZ and SPY-QQQ extend across zero.
Analysis 05
The stress effect fades with the horizon
The stress response is strongest over 20-30 trading days for the cross-market pairs. By 60 days, the increase is materially smaller. SPY-QQQ behaves differently. Its incremental response falls quickly because coupling is already structurally high. Stress-induced coupling is therefore strongest at shorter horizons and weakens as the measurement window lengthens.
Cross-market pairs show their largest mean increase at 20 days, with smaller increases at 30 and 60 days. SPY-QQQ shows its largest incremental increase at 10 days.
Analysis 06
Similar volatility does not mean shared shocks
The standalone volatility profiles do not map neatly onto the coupling results. SPY and FEZ are relatively similar across several parameters, yet SPY-QQQ remains the more persistently coupled relationship. QQQ also differs materially in mean reversion, long-run variance and vol-of-vol, while EEM stands apart on price-volatility correlation. Shared variance shocks therefore cannot be inferred from univariate volatility dynamics alone.
Standardised mean reversion, long-run variance, vol-of-vol and price-volatility correlation are respectively: SPY -0.98, -0.84, -1.07, -0.34; QQQ +1.67, +1.16, +1.45, -0.42; FEZ -0.25, -1.14, -0.77, -0.93; EEM -0.43, +0.81, +0.39, +1.69.
Methodology
How the analysis is constructed
- Sample period
- 03 January 2007-24 September 2026
- Stress reference
- VIX is used as the external market-stress reference.
- Data source
- Yahoo Finance, accessed through yfinance. Daily open, high, low and close prices are used to construct volatility measures. Incomplete current-day observations are excluded.
- Data through
- 24 September 2026
- Equity-market exposures
- SPY — S&P 500
- QQQ — Nasdaq 100
- FEZ — Euro Stoxx 50
- EWJ — Japan
- EEM — Emerging Markets
Volatility measure
Daily variance is estimated using the Garman-Klass OHLC estimator. Close-to-close and Parkinson variance are retained as robustness checks.
Heston-style variance dynamics
For each market, the variance process is estimated in Heston-style form using: mean reversion κ; long-run variance θ; vol-of-vol σ; price-volatility correlation ρ. Standardised variance innovations are then extracted from the fitted variance process.
Cross-market coupling
Pairwise variance-shock coupling is measured as the rolling correlation between standardised variance innovations. Core pairs: SPY-QQQ, SPY-FEZ, SPY-EEM and QQQ-EEM. The main specification uses a 20-trading-day window, with 10-, 30- and 60-day windows used for robustness.
Stress events
Major stress events are identified from local VIX peaks above the 95th percentile. Events are separated by at least 120 trading days to reduce overlap. Coupling is compared before, around and after each stress peak, with event-level bootstrap intervals used to assess uncertainty.
Research workflow
- Data
- Daily OHLC prices for SPY, QQQ, FEZ, EWJ and EEM
- Stress reference
- Daily VIX close
- Volatility measure
- Garman-Klass daily variance
- Robustness measures
- Close-to-close and Parkinson variance
- Variance model
- Heston-style mean-reverting variance process
- Parameters
- κ, θ, σ and ρ
- Variance shocks
- Standardised innovations from fitted variance dynamics
- Coupling measure
- Rolling correlation of standardised variance shocks
- Core window
- 20 trading days
- Robustness windows
- 10, 30 and 60 trading days
- Market pairs
- SPY-QQQ, SPY-FEZ, SPY-EEM and QQQ-EEM
- Stress events
- Local VIX peaks above the 95th percentile
- Event separation
- Minimum 120 trading days between stress peaks
- Event comparison
- Pre-event, peak and post-event coupling
- Uncertainty
- Event-level bootstrap intervals
Higher variance-shock coupling means that volatility innovations are moving more closely together across markets. The measure captures shared volatility shocks, not simply high volatility. A pair can be volatile without being tightly coupled, and tightly coupled without having identical standalone volatility dynamics. The event study compares coupling around separated VIX stress peaks. The results therefore describe how cross-market relationships change around major stress episodes. The results should be read as structural and event-based associations, not as estimates of a causal transmission path.
Research implications
What the evidence shows
The evidence in this report shows that volatility coupling is both structural and stress-dependent.
- 01SPY-QQQ has the highest average 20-day variance-shock coupling at 0.72.
- 02Cross-market pairs begin from lower coupling levels of around 0.58-0.60.
- 03Coupling rises sharply around major VIX stress peaks.
- 04The largest average increases occur in QQQ-EEM (+0.22) and SPY-EEM (+0.21).
- 05The stress effect is strongest over 20-30 trading days and weakens at longer horizons.
- 06Standalone Heston-style volatility parameters do not map directly onto shared variance-shock coupling.
Together, these results suggest that market stress does not affect every relationship in the same way. Some pairs already share volatility shocks in normal conditions, while others become substantially more coupled only when stress arrives. That distinction matters for cross-market risk. A relationship that appears relatively weak in calmer conditions can tighten quickly around major stress events. Volatility should therefore be read alongside the extent to which volatility shocks are becoming shared across markets.
Publication record
Sources and disclosures
Data
- Yahoo Finance, accessed through yfinance.
- SPY, QQQ, FEZ, EWJ, EEM and CBOE Volatility Index (VIX).
- Data through 24 September 2026.
Methodology reference
- Heston, S. L. (1993). “A closed-form solution for options with stochastic volatility with applications to bond and currency options.” Review of Financial Studies, 6(2), 327-343.
- Garman, M. B. and Klass, M. J. (1980). “On the estimation of security price volatilities from historical data.” Journal of Business, 53(1), 67-78.
- Parkinson, M. (1980). “The extreme value method for estimating the variance of the rate of return.” Journal of Business, 53(1), 61-65.
- Li, J.-C., Guo, J., Ma, R. and Zhong, G. (2025). “Dynamic forecasting and mechanisms of volatility synchronization in complex financial systems.” PLOS ONE, 20(10), e0334853.
- Hansen, P. R. and Lunde, A. (2005). “A forecast comparison of volatility models: does anything beat a GARCH(1,1)?” Journal of Applied Econometrics, 20(7), 873-889.
Software
Python, pandas, NumPy, SciPy, matplotlib and yfinance.
This publication has been prepared by LB Research for informational and educational purposes only. It does not constitute investment advice, investment research as defined under applicable regulation, a recommendation, or an offer to buy or sell any financial instrument.
The analysis is based on publicly available market data and the methodologies described within this publication. While reasonable care has been taken in preparing this report, no representation or warranty is made regarding the accuracy, completeness or timeliness of the information presented.
Any opinions expressed reflect the author's judgement at the publication date and may change without notice. Past performance is not indicative of future results.
Readers remain solely responsible for their own investment decisions.
© 2026 LB Research. All rights reserved. Questions or feedback: Contact LB Research
Download canonical PDFIn development: extending this research into cross-market structure and risk intelligence.