The Dollar Explains 2.6% of Emerging-Market Returns. It Is Still the Best Signal You Have.
Every emerging-market piece opens with the same claim: dollar down, EM up. Measured across 248 months of Federal Reserve and fund data, the direction holds and the strength does not — and for two years it ran the other way.
Directionally, yes. Measured from February 2006 to September 2026, months when the broad trade-weighted dollar fell produced a mean EM return of +1.56%, against -0.31% in months it rose — a spread of 1.88 points per month. But the correlation is only -0.16, meaning the dollar moves with about 2.6% of the variance in monthly EM returns. The relationship sorts months well on average and predicts any individual month badly.
Get new posts in your inbox
No spam, unsubscribe anytime.
About the Author
Sujit KarkiFinance Researcher & Market Analyst
Independent finance researcher and market analyst with expertise in macroeconomics, equity markets, and personal finance. I help regular investors make better-informed decisions through rigorous, data-driven analysis.
Measured over 248 months, the inverse dollar-EM relationship is real in direction and weak in strength: correlation -0.16, r-squared 0.026
Months when the broad dollar fell returned a mean +1.56% for EM against -0.31% when it rose — a 1.88-point monthly spread that compounds to roughly 20% annualized versus -4%
The rolling 36-month correlation ran positive from January 2016 through December 2017, so the relationship is not a constant and has inverted for years at a time
Most commentary quotes DXY, which is 58% euro by weight and a poor instrument for a claim about emerging markets; this uses the Federal Reserve broad trade-weighted index
The correlation is co-movement, not causation: both series respond to global risk appetite, and this measurement cannot separate them
Every emerging-market outlook opens the same way. A weakening dollar lightens
dollar-denominated debt, draws capital toward higher-yielding assets, and
steadies commodity prices — therefore emerging markets are set up to run. The
mechanism is plausible, it is repeated everywhere, and it is almost never
accompanied by a number.
An earlier version of this post did exactly that. It called the dollar "the
single most important macro driver" of EM returns and moved on.
That is a claim about two published series. It can be measured.
The claim everyone makes
The setup is straightforward: take the monthly total return of the longest-running
broad emerging-market fund a US reader can actually buy, take the monthly change
in the dollar, and see how much the two actually move together over twenty years.
One choice matters more than it looks. The dollar index most commentary quotes is
DXY, which is 58% euro by weight — a dollar-versus-Europe measure wearing a
general name. A claim about emerging markets needs the dollar against the
currencies emerging markets actually trade and borrow in. That is what the
Federal Reserve's broad nominal index is built for, so that is what this uses. It
costs about three years of history, since the broad index begins in 2006.
What 248 months show
The direction survives. The strength does not.
Correlation, EM returns vs the broad dollar0248 months, February 2006 to September 2026
An r-squared of 0.026 means the dollar moves with roughly two and a half
percent of the variance in monthly emerging-market returns. Ninety-seven percent
of what EM does in a given month is something else.
That is the number missing from every version of this argument, including the one
this post used to make. "The single most important macro driver" and "explains
2.6% of monthly variance" are not compatible statements.
And yet the relationship is not worthless — because it sorts months well even
though it predicts them badly:
Mean monthly EM total return by direction of the broad dollar, Feb 2006 – Sep 2026
Dollar direction
Months
Mean EM return
Annualized if sustained
Dollar fell
126
+1.56%
+20.4%
Dollar rose
122
-0.31%
-3.7%
Spread
—
1.88 points
—
Both things are true at once, and the tension between them is the whole point. A
1.88-point monthly spread is enormous — sustained, it is the difference between
compounding at twenty percent and slowly losing money. But an r-squared of 0.026
means you cannot use it to call any particular month. The dollar tells you which
weather you are in, not what tomorrow looks like.
The two years it ran backwards
A single correlation over twenty years can hide a relationship that moves. Here
it is on a rolling 36-month window:
Of 213 three-year windows, 177 were negative and 36 were positive. The
positive ones are not scattered noise — they cluster, and the longest run is
continuous from January 2016 through December 2017. For two full years, a
stronger dollar accompanied stronger emerging-market returns.
That is the part worth sitting with. Anyone who held the inverse relationship as
a rule during 2016 and 2017 was reasoning from a model the data had stopped
supporting, and would have had no way to tell from the commentary, because the
commentary does not carry the number that would have shown it.
The window has since swung back and is currently near its strongest: -0.33 as
of September 2026, against an all-time low of -0.46 reached in September 2025. The
relationship is real right now. It has also been real, and then not, before.
What this changes for an allocation
Three things follow, and none of them is "buy emerging markets because the dollar
is weak."
Treat the dollar as a regime indicator, not a trigger. The multi-year picture
is where the signal lives. At monthly resolution the r-squared says you are
reading noise.
Expect the relationship to lapse. It has spent 17% of its three-year windows
pointing the wrong way. A framework that only works while the correlation holds
needs to state what happens when it does not.
Size positions off volatility, not conviction in the mechanism. The
commonly repeated "5–20% of an equity allocation" is a convention, not a computed
result — this analysis does not produce a number to replace it, and it would be
the same failure to invent one.
What this does not mean
A correlation is not a mechanism. Both series respond to global risk
appetite. When investors retreat, they buy dollars and sell emerging-market
equities in the same move — which produces exactly this correlation without the
dollar causing anything. This measurement cannot separate the two, and nothing
above should be read as establishing that it can.
One fund is not the asset class. EEM tracks MSCI Emerging Markets, which is
heavily weighted toward a handful of large markets. A different EM index would
give a somewhat different number.
Twenty years is not forever. The sample opens in 2006 and therefore contains
the 2008 crisis, the taper tantrum, the pandemic, and the 2022 tightening cycle —
four dollar-spike events in 248 months. A sample starting elsewhere would weight
them differently.
Nothing here is a forecast. The measurement describes what has happened. It
carries no claim about what the dollar or emerging markets do next.
Methodology
Every figure above comes from two primary series. The computed dataset is
published with this post as
em-dollar-cycle.json; the script that fetches and
computes it is
scripts/em-dollar-cycle.mjs.
Data. Emerging-market returns are monthly adjusted closes for EEM (iShares
MSCI Emerging Markets ETF), so dividends are reinvested and the figures are total
returns rather than price returns. The dollar is FRED series DTWEXBGS, the
Federal Reserve's Nominal Broad U.S. Dollar Index, published daily and averaged
across each month — averaging rather than taking the last print keeps a single
volatile session from defining the month.
Period. February 2006 to September 2026, 248 usable months. The start is set
by the broad dollar index, which begins in 2006.
Consecutive months only. An observation is dropped unless the month before it
is the immediately preceding calendar month, so a gap in either series can never
be read as a one-month move. One month was dropped on this rule.
Correlation. Pearson, on monthly percentage changes in both series. The
rolling figure uses a 36-month trailing window, giving 213 windows.
Reproducing it.npm run em:dollar in the repository re-fetches from Yahoo
Finance and FRED and prints the current table. The figures above will age; the
command will not.
Disclosure. The commentary was drafted with AI assistance; the data
selection, the calculation, and the conclusions are mine.
Directionally, yes. Months when the broad trade-weighted dollar fell produced a mean EM return of +1.56%, against -0.31% in months it rose — a spread of 1.88 points per month. But the correlation is only -0.16, meaning the dollar moves with about 2.6% of the variance in monthly EM returns. It sorts months well on average and predicts any individual month badly.
Sujit Karki|||6 min read