How Harvard Researchers Accidentally Doubled the Tariff Effect
When Federal Reserve Chairman Jerome Powell mentioned rising import prices as proof that tariffs were driving inflation, he likely referenced new findings from Harvard Business School. Their key conclusion indicates that import prices have surged “roughly twice” as much as domestic prices—specifically, a ratio of 1.8 times—which suggests a significant impact of tariffs on traded goods.
Yet, there’s a significant caveat. This discovery is heavily reliant on two methodological choices that appear somewhat questionable, choices that, if altered, could reduce the observed effect by around 40%.
The study, titled “Tracking the Short-Term Price Impact of US Tariffs,” involves a team led by Alberto Cavallo from Harvard, along with researchers from Northwestern and San Andres universities. It’s certainly causing ripples in Washington for good reasons—being based on fresh, real-time pricing data addresses urgent policy issues. However, flaws within the study undermine the confidence we might have in the results.
Baseline Selection Techniques
To assess the tariff effect, researchers rely on a “pre-tariff trend” to indicate price directions prior to customs duties, with deviations measured after tariffs are imposed. It sounds straightforward enough.
But when does the baseline period start, and when does the “tariff period” begin? The Harvard team made key choices: they started the baseline in October 2024 and treated March 4, 2025, as when tariffs began. The rationale for starting in October isn’t explained. They do have data dating back to January 2024 tucked away in an appendix, but they don’t utilize it in the main analysis nor clarify their reasoning.
We conducted a test using publicly available data to see how sensitive the findings were to these choices. Based on their conclusions, import prices rose by 5.4%, while domestic prices increased by 3.0%, yielding a difference of 2.4 percentage points—a ratio reported as 1.8 times.
However, this finding hinges entirely on their specific choices. If we stretch the baseline to start in January, the difference between the price changes for imported and domestic goods narrows down to 1.6 times. Basically, that’s a 14% reduction in the gap by merely adjusting the baseline start date.
Why does this matter? The researchers had access to data from January onward but opted to commence from October. A more extended baseline captures more of the underlying trend, thus making post-tariff deviations appear smaller.
The choice of March 4 as the tariff start date is based on announcements related to major tariffs on China, Mexico, and Canada. It seems more natural to consider the date of Trump’s reelection—after all, he defeated rivals who campaigned against tariffs and argued that they functioned as a “national sales tax.”
If retailers project into the future—and the Harvard paper claims to do just this, focusing on how prices react to “tariff news” —expectations would likely start influencing prices from Trump’s election victory. If we use this approach, the ratio drops from 1.8 to 1.4, which translates to “about 40 percent more” rather than “about twice as much.”
What if we instead change the tariff date to April 2, the day Trump announced a 10% basic tariff on nearly all imports? This broad intervention leads to a different interpretation of the March price change as a “pre-tariff base value.” By maintaining the January baseline, we see the absolute difference drop to 1.59 times.
Do you notice a pattern? At every decision point, the choices made by the researchers tended to generate larger absolute effects and ratios. They picked a later starting reference period and an earlier treatment date. While each choice can be justified, when combined, they inflate the measured impact.
The Garden at the Fork in the Road
This situation resembles what sociologist Andrew Gelman refers to as a “Garden of Forking Paths.” Researchers, without fabricating data or hacking results, face decisions about timeframes, variables, and controls—all of which can yield different results.
The issue is not that these researchers made irrational choices; rather, every decision leaned toward maximizing the headline discovery. They didn’t test whether their findings hold under alternative specifications. Although their appendix displays data back to January 2024 framed as a “robustness check,” this doesn’t provide evidence that the main results hold up when baselines are modified.
Was it wrong to start in October? Not necessarily. Is March 4 an unreasonable treatment date? You could argue that. Yet making all these decisions in one direction without assessing their sensitivity while touting an “approximately 2x” result is another issue altogether.
A Black Box
The paper claims that tariffs raised overall inflation by 0.7 percentage points. But will this assertion withstand scrutiny? Unfortunately, we can’t be sure. The researchers base this finding on aggregated price indices and disconnected CPI weights. They measured “official CPI spending shares at the three-digit COICOP level,” stating that their sample encompassed 29.7 percent of the CPI basket, yet fail to report specific weightings or category-level contributions that sum to that 0.7 percentage point.
It’s essentially a black box. We see that the sample covers about 30% of the CPI, with furniture—showing the largest price rise—accounting for 54% of its products. However, it’s unclear whether furniture makes up 5% or 15% of the actual CPI weight. Without these details, the 0.7 percentage point claim cannot be verified or reproduced.
This lack of transparency becomes more concerning given the sensitivity of their other conclusions to seemingly questionable measurement choices.
This paper is likely to be cited in policy discussions as if its findings were scientifically rigorous. However, the main finding that imported goods have risen by “about twice as much” compared to domestic goods seems to be just one of several equally plausible interpretations.
If we change the baseline period to incorporate all available data, that “about 2x” becomes a “60 percent increase.” If we consider Trump’s election as the initiation point for tariff expectations influencing prices, the increase drops to “more than 40%.” Using April 2 as the tariff date reduces the absolute difference substantially.
This doesn’t imply that the researchers are being dishonest—in fact, it’s the opposite. They are a solid team engaging in serious research. Yet, it highlights how the flexibility afforded to researchers allows for numerous small choices in analysis that can yield results that, while citable and policy-relevant, may not hold under rigorous examination.
The next time you hear that imported goods are about twice as expensive as domestic ones due to tariffs, it might be worth asking about the standard used and the treatment date. Depending on that information, the impact might be significantly less than advertised.
