Direct answer: what evidence suggests rational expectations doesn’t hold in forex
Rational expectations in forex means that market participants’ forecasts of currency movements use available information in a way that is internally consistent with the probabilistic model generating prices. Evidence that rational expectations may not hold typically comes from forecast-performance and error-dynamics tests: researchers look for patterns where predictions do not incorporate information efficiently, or where changes in information do not lead to forecast updates that match what the model would imply.
Common indicators include (1) persistent forecast errors that do not become “unpredictable noise” after accounting for risk and model structure, (2) systematic correlations between past information and future errors, (3) imperfect calibration—forecasts that are not statistically consistent with realized outcomes—and (4) instability of estimated relationships as conditions change, which can indicate that the model participants use is incomplete or not shared.
Explanation: how these tests map to the rational expectations idea
Researchers typically treat rational expectations as a testable statement about how information should be reflected in forecasts and prices. If forecasts are rational under a correct model, then after conditioning on the information set, the remaining forecast errors should have desirable statistical properties (for example, no exploitable predictability using the same information that was assumed to be incorporated).
In forex, the difficulty is that “forecasts” can be defined in several ways: expected depreciation/appreciation from an exchange-rate model, inflation-expectation-based channels, or expectations embedded in survey data or market instruments. Evidence against rational expectations usually comes from mismatch between what the model implies and what is observed:
- Forecast errors are not independent of relevant variables included in the information set.
- Expected outcomes are consistently biased, even after accounting for risk premia and measurement choices.
- Parameter estimates and relationships change across time windows, suggesting that the effective model used for expectations is not stable.
Example checks: patterns researchers use to challenge or defend the hypothesis
Because forex data are noisy, single results are rarely decisive. Researchers therefore emphasize robustness checks, such as comparing multiple specifications:
- No remaining predictability: test whether variables available at the forecast origin still predict the sign or magnitude of later errors.
- Alternative conditioning sets: examine whether results survive when the assumed information set is expanded.
- Model comparison: check whether a simple rational-expectations-style model is outperformed by richer dynamics, while being careful not to confuse fit with true information processing.
- Heterogeneous risk and constraints: evaluate whether “failures” could instead be risk-premium movements, leverage/portfolio constraints, or volatility regimes—factors that can break the clean interpretation of expectations.
If observed patterns remain after these checks, they are often interpreted as evidence that expectations are not fully rational under the tested model assumptions. However, the interpretation is always conditional on how risk, information measurement, and model form are handled.
Limitations and uncertainty: what the evidence cannot prove
Several limitations matter:
- Rational expectations is model-dependent: you can reject “rational expectations” for a specific information set or probability model, without proving that no participant is rational.
- Forex is affected by multiple channels beyond expectations of fundamentals (including risk, liquidity, and regime changes), which can look like irrational updating if not modeled explicitly.
- Measurement uncertainty: different proxies for “expectations” (surveys, market-implied measures, or model-based expectations) may not align, so evidence can reflect proxy error.
- Statistical power and noise: even if expectations are mostly rational, finite samples may produce apparent predictability.
So, the most defensible conclusion is bounded: certain empirical regularities in forecast errors and information efficiency are consistent with rational expectations failing under specific assumptions, but they do not uniquely identify a single cause.