Factor Investing Group
Fig Research Suite · Beta Lab
Research Tools
Fig Beta Lab
Fig Beta Lab · Global Beta Evolution · Original Research
Beta Is Not a Number.
It Is a History.
It Is a History.
27 MSCI country proxies. 23 years of daily returns. Three rolling windows. One finding: the single beta number reported for any country to MSCI World is one of the most misleading statistics in global investing. This tool shows you the full distribution — how it evolved, how unstable it has been, and how differently every country’s relationship to global markets has structured itself over time.
Original research by Victor Melfa, portfolio manager · Factor Investing Group · Centre for Financial History, University of Cambridge · Johns Hopkins University (MS Finance, 2023) · A novel index-consistent cross-country beta evolution study.
Panel 01 · 23-Year Rolling Beta to MSCI World
Beta Evolution
Select any of the 27 countries. Toggle the rolling window. Watch how the relationship to global markets has lived — and changed. The window choice is itself a modeling assumption.
Rolling Beta Definition
βt(window) = Cov(rfund, rworld)t ÷ Var(rworld)t — recalculated daily over 126 / 252 / 504 trading day windows
Plain English: On any given day, beta answers: "over the past N trading days, how much did this country move for every 1% move in the world market?" A beta of 0.8 means it moved 0.8% on average for every 1% world move. But that number changes every single day as the window rolls forward. This tool shows you all those changes — not just today’s answer. Shorter window = more reactive but noisier. Longer window = smoother but slower to detect regime shifts. All three show the same underlying relationship at different levels of resolution.
Rolling window:
Select a country above
Select a country to see its beta evolution and statistics.
Panel 02 · Beta Distributions
The Distribution Is the Message
A single beta describes the mean of a distribution. These histograms show what that distribution actually looks like — wide, narrow, fat-tailed, or in some cases, with two distinct humps that suggest structural regime change.
Rolling window:
Select a country above
The shape of this distribution is more informative than any single number within it.
Panel 03 · All 27 Countries Ranked
No Two Countries Are Alike
All 27 MSCI country proxies ranked by mean beta, beta volatility, and distribution width. Click any column header to re-sort. Click any row to see that country’s full evolution.
Rolling window:
| Country ↕ | Market | Mean β ↕ | Volatility σ ↕ | Range ↕ | Bimodal? |
|---|
Reading the table
Mean β — the average beta over the full period. This is the number most investors use. It hides everything else in this table.
Volatility σ — standard deviation of the rolling beta series. High σ = highly unstable relationship to world markets.
Range — max minus min. A country with range 1.5 has been both near-zero and highly correlated at different points in the same 23-year period.
Bimodal — whether the distribution shows two distinct humps. This indicates structural regime change — not smooth drift.
Volatility σ — standard deviation of the rolling beta series. High σ = highly unstable relationship to world markets.
Range — max minus min. A country with range 1.5 has been both near-zero and highly correlated at different points in the same 23-year period.
Bimodal — whether the distribution shows two distinct humps. This indicates structural regime change — not smooth drift.
Panel 04 · Rolling Annual Correlation · 2002–2022
Correlations Are Not Stable Either
The same instability that defines beta also defines correlation. Move the slider through time to watch the entire correlation structure evolve. Some pairs tighten in crisis. Some decouple. No two years are identical.
How to read this chart
Each cell shows the pairwise correlation of daily returns between two countries over the trailing 12 months ending in the selected year. Deep green means the two markets moved closely together. White/light means near-zero correlation — independent movers. Red means they tended to move in opposite directions.
What to look for: Drag the year slider and watch how the entire matrix shifts. Notice which country pairs are persistently correlated, which decouple over time, and which show sudden changes. The diagonal is always 1.0 (each country correlates perfectly with itself).
Window choice: Shorter windows (1-month, 3-month) react faster and show more dramatic color shifts — especially during crises. Longer windows (1-year) smooth out noise and show structural relationships. The same modeling assumption as the rolling beta window. Neither is more correct. They show different time scales of the same relationship. Expect the most dramatic color and movement at 1-month — individual pairs can swing from near-zero to near-perfectly-correlated within a single calendar year. More stability at 1-year.
Color scale note: The neutral color (white) is set at 0.30 — the approximate average pairwise correlation across this dataset. Anything above 0.30 reads green; below reads yellow-orange-red. This is appropriate for equity data: a correlation of 0.55 between two country equity markets is genuinely elevated, not neutral.
On diversification: While individual country pairs show meaningful variation, holding a broad multi-country portfolio does smooth these relationships somewhat — total world funds tend to be more strongly and stably correlated with each other precisely because they are aggregating across many of these same pairs. The instability shown here is most relevant for concentrated single-country exposures.
What longer windows reveal: The data does show that pairwise correlations tend to stabilize as the rolling window lengthens — the 1-year view is visibly smoother than the 3-month view, and the structural relationships between markets (particularly within Europe) become clearer. This is not a guarantee of future stability. As the 3-month window shows, correlations can be extremely choppy in short periods, and the longer-window stability can unravel quickly when regimes shift. The window choice is a modeling assumption, not a truth. Just like the beta story, a single correlation number hides a complex history.
What to look for: Drag the year slider and watch how the entire matrix shifts. Notice which country pairs are persistently correlated, which decouple over time, and which show sudden changes. The diagonal is always 1.0 (each country correlates perfectly with itself).
Window choice: Shorter windows (1-month, 3-month) react faster and show more dramatic color shifts — especially during crises. Longer windows (1-year) smooth out noise and show structural relationships. The same modeling assumption as the rolling beta window. Neither is more correct. They show different time scales of the same relationship. Expect the most dramatic color and movement at 1-month — individual pairs can swing from near-zero to near-perfectly-correlated within a single calendar year. More stability at 1-year.
Color scale note: The neutral color (white) is set at 0.30 — the approximate average pairwise correlation across this dataset. Anything above 0.30 reads green; below reads yellow-orange-red. This is appropriate for equity data: a correlation of 0.55 between two country equity markets is genuinely elevated, not neutral.
On diversification: While individual country pairs show meaningful variation, holding a broad multi-country portfolio does smooth these relationships somewhat — total world funds tend to be more strongly and stably correlated with each other precisely because they are aggregating across many of these same pairs. The instability shown here is most relevant for concentrated single-country exposures.
What longer windows reveal: The data does show that pairwise correlations tend to stabilize as the rolling window lengthens — the 1-year view is visibly smoother than the 3-month view, and the structural relationships between markets (particularly within Europe) become clearer. This is not a guarantee of future stability. As the 3-month window shows, correlations can be extremely choppy in short periods, and the longer-window stability can unravel quickly when regimes shift. The window choice is a modeling assumption, not a truth. Just like the beta story, a single correlation number hides a complex history.
Year:
Window:
2008
Jump to:
<0 0.15 0.30 0.55 0.85+
1-year trailing correlation of daily returns
Move the slider or click Play to animate through time. Hover any cell to see the exact pair correlation.
Panel 05 · The Honest Conclusion
What the Data Actually Says
After 23 years, 27 countries, and 141,000+ daily beta observations — here is what this research establishes. And here is what it does not.
What this research demonstrates
1. Beta is not a parameter. It is a time series. Every country shows meaningful variation in its rolling beta. None are constant. Using a single full-period beta to represent a country’s current risk relationship to global markets is a simplification that borders on fiction.
2. Developed markets are not necessarily more stable. Japan’s mean beta is −0.025. Australia’s is 0.032. These are among the largest, most liquid equity markets in the world — and they have been near-zero beta to MSCI World for most of 23 years. The “developed = stable and correlated” assumption does not survive contact with this data.
3. Emerging market beta instability is quantifiably different. India (σ=0.417), Turkey (σ=0.276), Indonesia (σ=0.262) show beta volatility 2–3× higher than the most stable developed markets. The risk is not just in the level of beta — it is in how unpredictably it moves.
4. Some distributions have two humps — not one. India, Taiwan, Turkey, Sweden, Germany, Poland all show bimodal beta distributions. This is not noise. It suggests that the country’s relationship to global markets has fundamentally reorganized at some point in the sample.
2. Developed markets are not necessarily more stable. Japan’s mean beta is −0.025. Australia’s is 0.032. These are among the largest, most liquid equity markets in the world — and they have been near-zero beta to MSCI World for most of 23 years. The “developed = stable and correlated” assumption does not survive contact with this data.
3. Emerging market beta instability is quantifiably different. India (σ=0.417), Turkey (σ=0.276), Indonesia (σ=0.262) show beta volatility 2–3× higher than the most stable developed markets. The risk is not just in the level of beta — it is in how unpredictably it moves.
4. Some distributions have two humps — not one. India, Taiwan, Turkey, Sweden, Germany, Poland all show bimodal beta distributions. This is not noise. It suggests that the country’s relationship to global markets has fundamentally reorganized at some point in the sample.
The Fig Position on Regime Breaks
India’s bimodal distribution shows two distinct beta regimes clearly separated in the data. We don’t know why the break occurred. We don’t pretend to. The industry’s instinct — to identify the break, explain it, and build a forward model around it — is exactly the kind of false precision this research argues against. The data is the warning, not an invitation to model the break.
What this research establishes
Beta is a time series, not a number
Developed markets can be near-zero beta
Emerging market beta instability is measurably higher
Bimodal distributions exist and are meaningful
What this research does not claim
It does not predict future beta
It does not explain why regimes change
It does not recommend avoiding unstable countries
It does not remove the need for judgment
Research Note
27 MSCI country proxy funds: Switzerland, India, Mexico, South Korea, Taiwan, Frontier Africa, Canada, Japan, Netherlands, South Africa, Europe, Australia, Malaysia, Hong Kong, France, Italy, Germany, Denmark, Spain, Sweden, Indonesia, Singapore, Belgium, Poland, Turkey, China, and Pakistan. Daily return data spans January 2001 – December 2023 (~5,000 trading days per country). Rolling betas computed at 126-day (6-month), 252-day (1-year), and 504-day (2-year) windows against MSCI World as benchmark. Correlation matrix from 5,031 common trading days. Research by Victor Melfa, Centre for Financial History, University of Cambridge · portfolio manager, Factor Investing Group