X In April 2025, as the trade conflict escalated and American stocks sold off, the dollar behaved differently depending on when you looked. During the trading session, it still tended to rise against the euro in the same minutes that stocks fell. Over the following days, it weakened. Someone watching a five-minute chart and someone checking their portfolio a week later could reasonably have reached opposite conclusions about whether the dollar was providing shelter.
In my forthcoming paper, I separate those two responses. I call the first the dollar’s safe haven reflex and the second its persistence. The distinction turns out to be the main result: in 2025, the reflex survived, but the protection did not last. The evidence points to a turbulent year in which the dollar’s behavior changed, rather than a permanent end to its international role.
What It Means to Be a Safe Haven
A safe haven is an asset that rises when risk spikes. That is a more demanding property than simply remaining stable. A government bond can preserve most of its value during a stock market crash without gaining enough to offset losses elsewhere in a portfolio. It can remain a safe asset while failing to provide the particular protection investors expected.
The same distinction applies to currencies, with an additional complication: a currency’s value is always measured against another currency. In the main results below, the dollar means the dollar against the euro. I also compare it with five other currencies, because weakness against the euro alone would not establish a broader change in the dollar’s behavior.
There is also the question of time. The currency investors reach for immediately may differ from the one they choose to hold once they have had several days to reconsider. A single daily price change can hide that distinction.
The Data and the Two Clocks
I use futures prices from June 2010 through June 2026, covering stocks, currencies, government bonds and gold. The sample contains 465 days on which S&P 500 futures fell by more than 1%. These are the paper’s stress days, defined by an observable fall in stock prices rather than by a judgment about which news stories mattered.
For each stress day, I ask two questions:
-
What happens within the session? I compare five-minute changes in stocks and the dollar during U.S. trading hours. Does the dollar rise in the same minutes that stocks fall?
-
What remains several days later? I measure the dollar’s cumulative change from the close before the selloff to five trading days after it. Does a deeper stock market decline go together with a larger dollar gain, or a larger loss?
The second measure includes the selloff day itself. It is not the return to a trade placed after the decline has already happened. That matters when interpreting the numbers.
1. The Immediate Reflex Survived
In the recent years of the sample, the dollar still tended to rise against the euro in the same five-minute intervals in which stocks fell. The relationship survives adjustments for movements in Treasury prices and for the exact opening and closing times of the New York stock market. It is also stronger when I restrict the analysis to days with better intraday currency coverage.
This reflex is not an unchanging feature of the dollar. It disappeared for a stretch in the mid-2010s. But it was present again from 2021 through the first half of 2026, including the period in which the longer response went the other way. Whatever changed in 2025, it did not eliminate the dollar’s immediate bid against the euro when stocks fell.
2. Over the Following Days, the Relationship Reversed
Before 2024, the relationship between the depth of a stock market selloff and the dollar’s subsequent cumulative return was weakly protective. In the 2024–2026 window, it turned the other way. Each additional percentage point of stock market decline was associated with roughly three-quarters of a percentage point more dollar depreciation by five trading days after the selloff. The reversal is concentrated in 2025.
Average returns give a simpler sense of scale. Across the earlier stress days, the dollar gained about 0.28% over the measurement window. In 2024–2026, it lost about 0.31%. These are meaningful differences for an asset expected to cushion losses, but they are not the movements of a currency in collapse.
The pattern survives changing the threshold for a bad stock market day, using a basket of currencies, and checking Federal Reserve spot exchange rates instead of futures. Removing April 2025 roughly halves the estimated relationship but does not eliminate it. Defining stress through jumps in the VIX, the options market’s gauge of expected stock market turbulence, produces a smaller reversal. The size therefore depends on which kind of stress we measure.
A Caveat About How Much Evidence There Is
The recent comparison contains 66 stress days with complete return windows. Those are not 66 independent experiments. Several belong to the same run of bad news, and their following days overlap. Grouping them into separate stress episodes leaves 35, with the spring 2025 tariff turmoil carrying much of the result.
When I assess uncertainty by treating whole episodes as the independent pieces of evidence, the estimated recent reversal is no longer precise enough to rule out zero. The change from the earlier period does remain statistically clear. Those statements answer different questions: the evidence is stronger that the dollar’s response changed than that we know exactly how far its recent response lay on the wrong side of zero. The distinction is worth keeping even in a short account of the paper.
3. The Dollar Lost Most Ground to Other Havens
The estimated losses are largest against the Swiss franc, followed by the euro and the yen, and smallest against the Australian dollar. That ordering matters. The first three are competing reserve or haven currencies; the Australian dollar is more closely tied to commodities and the economic cycle. The pattern looks more like a change in where investors sought protection than a situation in which everything simply fell together. The individual estimates are uncertain, however, so I treat the ordering as suggestive.
German government bonds also show an estimated rally over the same horizon, although that result narrowly misses the more conservative test based on whole episodes. Two other findings limit the story:
-
The Treasury evidence is inconclusive. The estimates are too imprecise to establish whether long-term U.S. government bonds retained their protection over several days. Their behavior during one memorable April selloff cannot settle the question for the full sample.
-
Gold depends on the currency used to measure it. Gold rose in dollar terms, but its response was small and statistically unclear when measured in euros. Much of the apparent gold rally was the other side of a weaker dollar.
These are comparisons of returns. They do not track money moving from one investor’s dollar holdings into that same investor’s German bonds.
4. The Options Market Priced the Same Concern
Options provide a separate check because they price protection against future moves. I compare the relative price of protection against dollar weakness and dollar strength, using options on euro futures. In 2024, the pricing tilted toward dollar strength. In 2025, it tilted toward dollar weakness, with the turn appearing in the second quarter, during the tariff turmoil. By spring 2026, the tilt had reversed again.
Options against the Swiss franc show a similar increase in the relative price of protection against dollar weakness in 2025. Options against the yen do not. The evidence therefore has limits, but two separate currency options markets reinforce the return evidence: dollar downside became more expensive to insure against during the year in which its protection failed to persist.
Why the Two Responses Might Diverge
One interpretation is that immediate liquidity needs and slower portfolio decisions pull in different directions. Investors facing sudden stress need somewhere to put cash, and dollar markets remain deep and easy to trade. Once the immediate pressure passes, they can reconsider where they want to hold it. If the source of their concern is U.S. policy itself, a competing haven may look more attractive. A simple model in the paper shows how those two decisions can produce the pattern in the data.
The reversal is concentrated on days without a scheduled Federal Reserve decision, payrolls report or inflation release. On the small set of scheduled release days, the estimate is close to zero. That fits the interpretation, but does not establish it: unscheduled news also includes foreign crises and other financial events, and the difference between the two groups is too imprecisely estimated to be conclusive. The paper measures how prices moved together, not a causal effect of lost confidence in U.S. policy.
What 2025 Does and Does Not Tell Us
The reversal fades in the first half of 2026. The dollar’s estimated response moves back toward protection, although there are too few stress days to establish a durable recovery. The options evidence also turns back. Calling the 2025 pattern permanent de-dollarization would therefore go beyond what the data show.
The lesson is narrower and useful: whether a currency provides shelter depends on how long that shelter is needed. In 2025, the dollar could still attract an immediate bid against the euro while failing to protect over the days that followed. Assessing its haven role requires checking both horizons. The first few minutes did not tell the whole story.
References
-
Torul, O. (2027). Reflex without Persistence: The Dollar’s Multi-Day Safe Haven Reversal in 2025. Journal of International Financial Markets, Institutions & Money. 114, 102413.
-
Replication code and derived data are deposited at Harvard Dataverse.