This is the correlation lens that traditional diversification is built on, just to serve as some summary of what I'm talking about - the US is pinned to the VIX in most regimes with Canada and Mexico following. Europe generally sits between -0.34 and -0.42 over the full sample. The emerging markets usually get sold as "diversifiers" are pulled in during the crisis periods (2008 and 2020). Mexico goes to -0.63 and Brazil to -0.61 in 2007-2009. Asia sits near zero, and maybe some of that is due to timing, even after lagging all the Asian closes by a day.

So what? What's surprising to me is that post-Covid, the European correlations loosened a lot (as seen by Spain going to -0.18 and Italy going to -0.24) while Japan moved to -0.25, which was its strongest reading in any regime. The currency side is flat, with no currency clearing +- 0.14 in any regime. If correlations were the story, seems like FX can't help us with any information on equity fears (seems like otherwise from a spillover perspective).

Total connectedness is the average share of each market's 10-day forecast-error variance that comes from shocks in other markets (the Diebold-Yilmaz "from others" measure). It rises pretty clearly in both crises, as you'd expect. Equities going from 72% to 89% in 2007 and then back to 86% in Covid, with the QE decade in between.

Well, the bigger finding in my opinion is that currency connectedness didn't come down after Covid, and it hit 84% in 2023-2024. That's the highest anywhere in the sample, and it matches the equity picture. Rate hikes, dollar strength and a lot of geopolitical noise hit currencies all at once. So, it's a more connected market, which is bad news if you count holding international currencies as a diversifier.

Net spillover is what a market sends out minus what it receives, so red markets are exporting volatility and blue markets absorb it. In equities, the core of Europe transmits persistently: France (+47), Germany (+43), the UK (+37), and the Netherlands (+31) over the full sample. Some of that is structural, since eight tightly linked European indices in a 25-market system do a lot of the talking to each other. The persistent absorbers are Indonesia, Japan, Saudi Arabia, Turkey, Taiwan, and China, with Japan absorbing 60 points more than it sends during the GFC.

The US doesn't really fit here. It was a strong transmitter before and after the GFC (+25 and +19) with its "to others" spillover peaking at 106% in 2007-09. This came down to neutral through QE and Covid, then came back up during 2023-2024. In the currency department, the Yuan was one of the biggest absorbers for 20 years before flipping to +21 post-Covid, level with the Euro as the largest transmitter. The Franc was a big transmitter before 2007 at +29, but has become a receiver since, while the Yen only turns positive in the GFC and in 2023-2024. Looks like transmitter status is NOT a fixed property of a market, so the screen had to be re-estimated for each of the five regimes I studied (this is a big problem if you extrapolate this to more and more granular data - how is correlation going to keep up against volatility spillover?).

Each dot here is the Sharpe ratio of a minimum-variance portfolio. Grey uses every market, blue only keeps the markets below the median spillover for all markets, and red keeps the ones above it. The dark columns are the two crises, and that's the important part. During the GFC, every screened portfolio beats the unscreened benchmark across all four designs in the 2x2 structure. Equities improve from -0.96 to somewhere between -0.62 and -0.87, and the currency portfolios go from -0.65 to basically flat (-0.05 and -0.08).

Equities-on-equities and currencies-on-currencies is important, but the cross-asset result is what matters most (equities-on-currencies and currencies-on-equities). On the low side, screening equity portfolios on currency spillovers (EC) beats screening them on their own spillovers (EE) in the pre-GFC window, the GFC, and Covid. On the high side, it also wins pre-GFC and in the GFC. FX seems to price shifts in global risk before equities absorb them, which makes it a better early-warning signal for equity fragility than equity signals themselves.

The calm periods are where things get harder. When nothing is breaking, the unscreened portfolio wins (makes sense too). Post-Covid equities earn 0.86 unscreened against 0.49 and 0.43 screened, because throwing out half the universe costs you the diversification you would normally get for free. Over the full 25 years it roughly washes for equities (0.24 vs. 0.18-0.24). That makes spillover screening a useful regime tool that complements the correlation-based allocation rather than replacing it.

Here's the full path of every case-study portfolio, shown as a median with its interquartile range. Each case study takes one country's spillover relationships against the rest, which gives 24-25 portfolios per design. The high-spillover equity portfolios end higher of course, at $331 and $381 per $100 against roughly $262 for the low side. This is because the markets that transmit volatility carry the most equity risk premium.

The low side falls less, exactly when it matters. It's median path had the shallower drawdown in both crises for all four strategies. In the GFC that's -47% vs -55% for EE and -44% vs -59% for EC, and in Covid its -32% vs -37% and -33% vs -41%. On the currency side, the low screen on equity spillovers held its GFC drawdown to 2%, against 12% on the high side. The currency sleeves mostly drift lower in dollar terms after 2011, which says more about long dollar bulls!

Lastly, how the eight strategies return streams move together. Over the full sample, they cleanly split into two groups. The equity strategies correlate to 0.79-0.91 with each other and are almost unrelated to the currency strategies (-0.13 to +0.16). So, the asset you end up holding matters more than the spillover strategy you screen on, which is to be expected when you are contrasting equities and currencies. Before 2007 everything was positively correlated, with even equity-vs-currency pairs running 0.26-0.65. That link mostly disappears after the GFC.

The crises pull the low and high sides apart from each other: The two equity-on-equity portfolios go from 0.71 correlated pre-GFC to 0.38 in the GFC, 0.05 in Covid and -0.14 post-Covid. In currencies, low and high spillover end up as somewhat mirror images (CCL vs CCH at -0.87 in Covid and -0.89 after). The screen doesn't just lower volatility, it changes which direction the shock goes. High-spillover portfolios tilt into systemic risk while low-spillover ones play defense, which is what makes this useful as a hedge.