How Economists Forecast Inflation and Currency Values — and Why They Often Miss

Inflation and exchange-rate forecasts shape decisions from central-bank interest-rate meetings to household budgeting, yet both remain among the hardest things in economics to predict accurately. Central banks, government statistical agencies and academic researchers have each documented, in their own technical publications, why this is the case.

How inflation forecasts are built

According to the Federal Reserve Bank of Richmond, forecasters start from price indexes such as the Personal Consumption Expenditures (PCE) index and the Consumer Price Index (CPI), including "core" versions that strip out volatile food and energy prices. On top of these, the Federal Reserve Bank of Cleveland notes that policymakers track specialized variants such as the Dallas Fed's trimmed-mean PCE and the Atlanta Fed's sticky-price CPI, which are designed to isolate price changes that are more likely to persist.

A common technique is what the Richmond Fed calls "signal extraction": statistical models that try to separate a lasting inflation trend (the signal) from short-lived price swings (the noise). As one Richmond Fed economist put it, if several inflation readings move in the same direction, a model will treat that as a sign of real underlying inflation rather than a temporary blip.

The Richmond Fed's research also points to a practical complication: statistical approaches tend to work reasonably well for forecasting inflation two or three months ahead, but they perform poorly when the economy undergoes a "big regime change" — a sudden shift in the forces driving prices, such as a pandemic-related supply shock.

Surveys and market prices as a check on models

Because no single statistical method is considered definitive, the Cleveland Fed explains that policymakers also consult survey-based measures of what people expect inflation to be. These include the University of Michigan Surveys of Consumers, the Philadelphia Fed's Survey of Professional Forecasters, the New York Fed's Survey of Consumer Expectations, and a survey of business executives known as the Survey of Firms' Inflation Expectations.

Market-based measures offer a third input: yields on Treasury Inflation-Protected Securities (TIPS) can be compared with ordinary Treasury yields to infer what investors expect inflation to average over a given period. The Richmond Fed's research found that, since 1980, statistical model-based forecasts have on average run about 0.5 percentage points lower than survey-based forecasts, and that statistical models reacted faster to an inflation run-up between 1998 and 2008 than surveys did.

The Cleveland Fed states plainly that "there is no consensus about which approach is better," which is why central banks typically weigh several measures side by side rather than relying on one.

Where inflation forecasts have gone wrong

Forecast errors are not merely theoretical. The Richmond Fed's account of the 2021 inflation surge notes that in April 2021 the Federal Reserve chair described a sustained rise in inflation as unlikely, but by that September officials were acknowledging that supply-chain bottlenecks had turned out to be larger and more persistent than previously anticipated. PCE inflation went on to reach 4.4% year over year by September 2021, according to the same research. The Richmond Fed also points to the 1970s, when U.S. inflation expectations are described as having drifted upward, largely undetected by contemporary models, starting from 1967.

How economists approach currency forecasting

Exchange-rate forecasting draws on a different toolkit, but it runs into an even starker limitation. One widely used theoretical anchor is purchasing power parity (PPP), described by the Federal Reserve Bank of St. Louis as an extension of the "law of one price": in theory, after adjusting for exchange rates, an identical basket of goods should cost the same in any country.

The St. Louis Fed illustrates this with a comparison of Starbucks latte prices: a latte priced at $7.65 in New York and 11 Swiss francs in Zurich implies a PPP exchange rate of 1.44 francs per dollar. The actual market rate at the time was 0.86 francs per dollar, meaning the franc traded far stronger than the simple price comparison suggested. The St. Louis Fed attributes this gap to several factors: many services such as haircuts or lattes cannot be traded across borders, PPP and actual exchange rates track each other only over long stretches of decades, and currency markets are also driven by capital flows, interest-rate differences and investor sentiment that price-level comparisons do not capture.

The random-walk problem

Beyond PPP, economists have tested many statistical models against a deliberately simple benchmark: the random walk, which simply predicts that tomorrow's exchange rate will equal today's rate. A 2026 study on the CAD/USD exchange rate, posted on the arXiv preprint repository, tested five statistical and machine-learning approaches — linear regression, random forest, gradient boosting, XGBoost and AdaBoost — against the random-walk benchmark. Only one method, the linear regression, achieved a statistically significant edge over the random walk. The study's authors found that even its best-performing models were dominated by very recent price history rather than broader economic drivers, a pattern they described as consistent with the near-random-walk behavior of exchange rates.

What this means for readers of inflation and currency news

The common thread across both fields is that forecasts are probabilistic estimates built on incomplete models, not guarantees. Inflation forecasts tend to be more reliable at short horizons of a few months and under stable conditions, but they can miss sharp turning points driven by shocks such as supply disruptions. Currency forecasts face an even tougher constraint: most published research finds that over short-to-medium horizons, theory-based and statistical models struggle to beat the simple assumption that tomorrow's rate will resemble today's.

• Inflation forecasts combine price-index data (CPI, PCE), statistical signal-extraction models, and surveys of consumers, businesses and professional forecasters.
• Central banks track multiple inflation measures at once because, per the Cleveland Fed, there is no agreed single best method.
• Currency forecasting relies partly on theories like purchasing power parity, which the St. Louis Fed says holds only over long time horizons, not short-term moves.
• Across decades of academic testing, most exchange-rate models still struggle to consistently outperform the random-walk benchmark, according to Fed-published research and a 2026 academic preprint study.