Dollar’s purchasing power “cut in half”? AI model challenges official US inflation statistics

An independent data project has used artificial intelligence to reconstruct price indices, arriving at conclusions that differ greatly from official data, raising profound market doubts about the true state of the U.S. economy.

An independent research project called RealityIndex shows that since the outbreak of the COVID-19 pandemic in 2019, the actual loss in dollar purchasing power may be as high as 50%, far exceeding the 26% announced by the U.S. Bureau of Labor Statistics (BLS).

The project is led by Tom Elliott, an independent data researcher residing in Madrid, who uses artificial intelligence tools to re-integrate BLS’s original price data, removing the effects of methodological adjustments made over time in official CPI calculations to restore the "unvarnished" price trajectory.

The implications of this conclusion for the market are significant. If the RealityIndex price index is used instead of official inflation data for GDP deflation, the U.S. economy has been in negative growth almost continuously since the summer of 2022, with just three quarters barely showing positive growth. Based on this, Grok AI estimates that compared to 2019, the cumulative real GDP loss in the U.S. ranges between 5% and 12%—meaning the so-called "post-pandemic recovery" may never have actually happened in terms of real data.

Official CPI: Eight Methodological Adjustments in 35 Years

The core argument of the RealityIndex project is that the U.S. official CPI has, through decades of methodological evolution, systematically underestimated actual price increases.

In 1983, BLS replaced direct housing prices with "Owner's Equivalent Rent" (OER), estimating rent based on homeowners hypothetically renting out their own homes instead of using actual mortgage and property tax payments. When home prices and mortgage rates rise faster than rents, this formula underestimates the inflation actually borne by households.

In 1996, the Boskin Commission lowered CPI based on the reasoning that consumers substitute products. Hedonic adjustments were introduced in 1998, suppressing price increases by attributing improvements to product quality—but notably, this adjustment never works in reverse; it does not increase prices when quality declines. In 1999, the geometric mean formula replaced the arithmetic mean, further masking increases in medical service costs.

The "chained CPI" was introduced in 2002, dynamically adjusting basket weights according to actual purchasing behavior. In 2018, hedonic adjustments were expanded to smart phones, internet services, cable TV, and many new categories. In 2024, BLS stopped tracking actual costs of medical services, only reporting insurance claims. In 2025, there were month-long data gaps.

These cumulative adjustments over 35 years have formed the systemic biases pointed out by Tom Elliott.

Data Gap: Actual Inflation Underestimated by 32% Since 1980

RealityIndex’s quantitative conclusions are alarming.

According to the index, a basket of goods worth $100 in 1980 will require $515 in 2025; whereas the official CPI puts the figure at only $391. This means actual price increases over the past 45 years have been 32% higher than official data. Extending the observation window to 55 years, the deviation increases to 54.4%.

Measured by purchasing power loss: official CPI shows that $1 in 1980 is now worth only 26 cents; RealityIndex shows its actual value has fallen to 19 cents.

Focusing on the six years since the pandemic, the gap is even more pronounced. Official data records a 26% loss in purchasing power, while RealityIndex estimates the loss is close to 40% to 50%. Regarding annual inflation rate trends, official CPI shows a peak in 2022 with a subsequent decline; RealityIndex shows that inflation never dropped below 6%, and the disinflation process hasn't proceeded as smoothly as official data suggest.

GDP Reassessment: Post-Pandemic "Recovery" May Be a Statistical Illusion

When the corrected price index is used as the GDP deflator, the U.S. economic picture fundamentally changes.

By official definition, an economic recession is two consecutive quarters of negative real GDP growth. Using RealityIndex data, since the summer of 2022 most quarters have seen real U.S. GDP in negative territory, with only three quarters just above zero. Grok AI estimates a cumulative GDP loss of 5% to 12% since 2019.

As a reference, losses during the Great Depression are considered one of history’s most severe cyclical shocks. The above estimates suggest that the cumulative economic loss from the pandemic and lockdowns may equal about half of the losses sustained during the entire Great Depression.

The article’s author also points out that in official GDP data, medical subsidies, government spending, and social services contribute a substantial proportion. If these factors are removed, the real performance of private sector economic activity would be even weaker. Labor force participation rate and employment-population ratio have never fully recovered since 2020 and continue to decline, confirming the above conclusions.

Methodological Debate and Limitations

The methodology of RealityIndex is not without controversy. The project’s basic logic is: directly use the original price data published by BLS, without incorporating any methodological adjustments, to construct a "pure" price index. Supporters believe this restores the actual price reality faced by consumers; critics may argue that some adjustments (such as hedonic corrections for quality improvements) have legitimate economic rationale, and removing them entirely may not be closer to reality.

Previously, the Brownstone Institute commissioned E.J. Antoni and Peter St. Onge in 2024 to conduct a similar study—which also concluded the U.S. has been in a technical recession since 2022—but this was strongly criticized by mainstream economists, who generally cited official data to question its conclusions.

Tom Elliott himself acknowledges that this project is still in the developmental stage and charting does not support real-time updates. The article’s author believes the method is replicable, and anyone can evaluate its methodology and raise objections. This is an important advantage compared to the closed official statistics system.

Another Interpretation of Depressed Consumer Confidence

These data may provide a new explanatory framework for a long-standing paradox that has puzzled analysts: Why is consumer confidence at historic lows when official data shows sustained economic growth?

If RealityIndex’s estimates are close to reality, then consumer pessimism is not "irrational" but a rational response to the sustained erosion of real purchasing power.

Citing data, the article notes that about a quarter of professional workers have had no nominal wage growth in the past five years; meanwhile, the proportion of dual-income households has risen from about one in five in 1950 to three in five today, meaning the labor input necessary to maintain the same living standard has actually increased substantially—a structural change not reflected in official household income statistics.

For investors, if actual inflation has been systematically underestimated, then real interest rates, real wage growth, and real GDP growth deflated by official CPI may all be systematically overestimated, warranting a reevaluation of the reliability of the macro benchmark data underpinning asset pricing models.

Risk Warning and DisclaimerThe market involves risks; investment should be cautious. This article does not constitute personal investment advice and does not take into account the individual investment objectives, financial circumstances or needs of any specific user. Users should consider whether any opinions, views or conclusions herein are suitable for their circumstances. Invest accordingly at your own risk.