{
  "id": "rational-expectations",
  "title": "Rational Expectations",
  "lang": "en",
  "source_lang": "en",
  "summary": "Rational expectations is the hypothesis that economic forecasts are consistent with the model governing outcomes and use available information without systematic, predictable errors.",
  "category": "economics",
  "subcategory": "macroeconomics",
  "type": "concept",
  "aliases": [
    "Rational expectations hypothesis",
    "Model-consistent expectations"
  ],
  "generated_at": "2026-11-04",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/rational-expectations/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/rational-expectations/",
      "json": "https://aiwiki.page/zh/rational-expectations.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Rational expectations is a hypothesis about how people forecast economic outcomes under uncertainty. It holds that their expectations are consistent with the economic relationships generating those outcomes, conditional on the information available to them. Forecasts need not be correct in every instance: unexpected events still occur. The central restriction is that forecast errors cannot be systematically predicted using the information on which the forecasts were based.\n\nIntroduced by John F. Muth in 1961 and developed extensively in macroeconomics during the 1970s, the hypothesis changed the analysis of inflation, economic fluctuations, financial prices, and public policy. When households and firms anticipate government actions, their responses can alter the effects of those actions. Expectations therefore cannot always be treated as a fixed extrapolation from the past.\n\nRational expectations is an assumption about beliefs, not a complete theory of economic behavior. It does not by itself imply perfect information, perfectly competitive markets, flexible prices, or ineffective stabilization policy. Its implications depend on the surrounding model: the information people possess, the constraints they face, how prices adjust, and the rules governing policy."
    },
    {
      "heading": "Meaning and mathematical foundations",
      "text": "The standard formulation identifies a person’s forecast with a conditional expected value. Let \\(x_{t+1}\\) be an economic variable observed next period, and let \\(\\mathcal I_t\\) denote the information available when a forecast is made at time \\(t\\). Rational expectations requires\n\n\\[x^e_{t+1|t}=\\mathbb E[x_{t+1}\\mid\\mathcal I_t].\\]\n\nThe expectation on the right is calculated using the probability law implied by the model. Consequently, the hypothesis links subjective forecasts to the model’s account of how the economy actually operates. In a fully specified rational expectations equilibrium, beliefs and outcomes must be mutually consistent.\n\nDefine the forecast error as\n\n\\[u_{t+1}=x_{t+1}-\\mathbb E[x_{t+1}\\mid\\mathcal I_t].\\]\n\nThe defining restriction then implies\n\n\\[\\mathbb E[u_{t+1}\\mid\\mathcal I_t]=0.\\]\n\nFor any suitable variable \\(z_t\\) contained in the information set, it also implies \\(\\mathbb E[u_{t+1}z_t]=0\\), provided the relevant moments exist. This orthogonality condition gives the hypothesis empirical content: information already available at the forecast date should not systematically explain subsequent forecast errors. The condition does not require errors to be normally distributed or to have constant variance.\n\nFor example, suppose inflation follows\n\n\\[\\pi_{t+1}=a+\\rho\\pi_t+\\varepsilon_{t+1},\\]\n\nwhere the parameters are known and the disturbance has a conditional mean of zero. If current inflation is observed, the rational forecast is \\(a+\\rho\\pi_t\\). Simply predicting that inflation will equal its latest value is consistent with this process only under particular parameter values, such as \\(a=0\\) and \\(\\rho=1\\). Rational expectations therefore does not prescribe one universal forecasting rule. The appropriate rule changes with the underlying economic process.\n\nThe same logic explains why rational expectations is not perfect foresight. A drought, an unexpected productivity innovation, or an unanticipated policy decision can produce a large forecast error without violating the hypothesis. Conversely, small errors are not necessarily rational if a forecaster persistently ignores information that would improve predictions. Under squared-error loss, the conditional mean minimizes expected forecast loss; under other loss functions, an optimal reported forecast may instead be a conditional median or another feature of the probability distribution.\n\nThe information set is essential. Two people can rationally make different forecasts if they possess different information. A firm observing its own orders may predict industry demand differently from an outside analyst. Rational expectations imposes consistency conditional on what each observes; identical forecasts follow only when relevant information and the assumed model are shared. Stronger assumptions of common information and common knowledge are additional modeling choices, not consequences of the hypothesis alone."
    },
    {
      "heading": "Origins and the transformation of macroeconomics",
      "text": "John F. Muth introduced the hypothesis in “Rational Expectations and the Theory of Price Movements,” published in Econometrica in 1961. His immediate concern was the formation of expectations in markets where production decisions precede the realization of prices. Agricultural and inventory examples made the problem concrete: expected future prices influence current production, but that production subsequently affects the prices being forecast. Expectations and market outcomes therefore have to be determined together.\n\nMuth proposed that forecasts should be consistent with the relevant economic theory rather than imposed as an arbitrary mechanical rule. The proposal challenged common uses of adaptive expectations, under which forecasts adjust gradually in response to past errors. Adaptive rules can be reasonable descriptions in some environments, but their parameters need not remain appropriate when policy or the behavior of the economy changes. Muth’s argument was not that every individual performs sophisticated calculations; it was that systematic, economically relevant forecasting errors require an explanation.\n\nRobert E. Lucas Jr. brought the approach into macroeconomic analysis during the early 1970s. His 1972 paper “Expectations and the Neutrality of Money” examined an economy in which producers observe local prices but cannot immediately distinguish changes in relative prices from changes in the general price level. Monetary disturbances can then affect output because people temporarily misinterpret the information conveyed by prices, even though their inferences are rational given what they observe. This was an information-based explanation of monetary non-neutrality, rather than an assumption that people repeatedly make the same avoidable mistake.\n\nThomas J. Sargent and Neil Wallace subsequently used rational expectations to analyze monetary policy and the distinction between anticipated and unanticipated policy changes. Their work helped establish new classical macroeconomics, which combined model-consistent expectations with optimizing behavior and, in influential early models, market clearing. The resulting conclusions challenged policy analyses that relied on a stable, exploitable relationship between inflation and unemployment.\n\nThe intellectual context included the expectations-augmented Phillips curve associated with Milton Friedman and Edmund Phelps. Their arguments distinguished temporary employment effects of unexpected inflation from the absence of a permanent inflation–unemployment trade-off. Rational expectations sharpened the question of how long predictable policy could continue to surprise economic agents. It also changed the interpretation of statistical relationships estimated under a particular policy regime.\n\nThe approach later spread beyond new classical economics. Real business cycle theory used rational expectations in models emphasizing technology and other real disturbances. New Keynesian economics combined it with imperfect competition, sticky prices, and other frictions. Lucas received the Nobel Memorial Prize in Economic Sciences in 1995 for his work on rational expectations and macroeconomic analysis; Sargent shared the 2011 prize with Christopher Sims for research on cause and effect in the macroeconomy."
    },
    {
      "heading": "Equilibrium, forward-looking models, and solution methods",
      "text": "Rational expectations becomes especially important when today’s choices depend on tomorrow’s outcomes and today’s choices also help determine those outcomes. A firm invests according to expected future sales; investment changes productive capacity; capacity influences future prices and profits. The resulting problem is not merely to forecast a fixed external process. It is to find an equilibrium in which the forecasts used to make decisions agree with the outcomes generated by those decisions.\n\nA simple forward-looking equation illustrates this structure:\n\n\\[y_t=a\\,\\mathbb E_t[y_{t+1}]+b z_t,\\]\n\nwhere \\(\\mathbb E_t\\) abbreviates conditioning on current information and \\(z_t\\) is an observed fundamental variable. Iterating the equation forward yields\n\n\\[y_t =b\\sum_{j=0}^{T}a^j\\mathbb E_t[z_{t+j}] +a^{T+1}\\mathbb E_t[y_{t+T+1}].\\]\n\nIf the terminal term converges to zero, the current value is a discounted sum of expected future fundamentals. This reasoning appears in models of asset prices, consumption, exchange rates, and inflation. Expectations about distant events can therefore affect current decisions before those events occur.\n\nA common solution technique is the method of undetermined coefficients. Suppose \\(z_t\\) follows \\(z_{t+1}=\\rho z_t+\\varepsilon_{t+1}\\), with a conditionally mean-zero innovation. Guessing a solution \\(y_t=c z_t\\) and substituting it into the original equation gives\n\n\\[c=a c\\rho+b, \\qquad c=\\frac{b}{1-a\\rho},\\]\n\nwhen the denominator is nonzero. The proposed forecasting relationship must reproduce itself after agents’ decisions are aggregated. This fixed-point requirement is the operational meaning of model consistency.\n\nLarger models contain both predetermined variables, such as inherited capital, and forward-looking variables, such as asset prices. Linear solution methods examine whether the system has the appropriate number of stable and unstable roots to determine a unique non-explosive equilibrium. Olivier Blanchard and Charles Kahn’s 1980 analysis supplied widely used conditions for linear models. Depending on the system, there may be a unique equilibrium, no admissible equilibrium, or multiple equilibria. Rational expectations alone does not guarantee existence or uniqueness.\n\nMultiplicity also leaves room for sunspot equilibria, in which an otherwise irrelevant signal coordinates beliefs and thereby influences real outcomes. Likewise, some asset-pricing equations admit bubble components unless additional restrictions exclude them. These possibilities show why rational expectations is not synonymous with a uniquely determined “fundamental” outcome.\n\nModern dynamic stochastic general equilibrium models commonly implement rational expectations through numerical methods. Researchers approximate decision rules near a steady state or solve nonlinear systems across a range of possible states. The distinction between the theoretical equilibrium and its computational approximation matters: linearization may be adequate for small fluctuations but less suitable around binding borrowing constraints, large shocks, or an effective lower bound on nominal interest rates."
    },
    {
      "heading": "Monetary policy, anticipation, and the Lucas critique",
      "text": "One of the best-known applications concerns the difference between expected and unexpected monetary policy. Consider a simplified supply relationship,\n\n\\[y_t-\\bar y_t =\\alpha\\bigl(p_t-\\mathbb E_{t-1}[p_t]\\bigr)+v_t,\\]\n\nwhere \\(y_t\\) is output, \\(\\bar y_t\\) is its normal level, \\(p_t\\) is the price level, and \\(v_t\\) is a supply disturbance. Output rises above normal when the price level exceeds what producers previously expected. If a monetary expansion is predictable from information available at \\(t-1\\), its price effects enter the expectation term and do not create the same surprise.\n\nIn models with this structure, a systematic policy rule cannot repeatedly raise output by generating predictable inflation surprises. This is the central intuition of the policy-ineffectiveness proposition associated with Sargent and Wallace. Its force depends on the model’s supply structure, information timing, and assumptions about wages and prices. It is not a theorem that follows from rational expectations in every possible economy. Nor does it imply that monetary institutions are irrelevant for inflation, volatility, or welfare.\n\nThe Lucas critique, developed in Lucas’s 1976 “Econometric Policy Evaluation: A Critique,” made a broader argument. Statistical relationships estimated under one policy regime may not remain stable after the regime changes. If households expect taxes to follow one pattern, their consumption behavior may differ from behavior under another pattern. Similarly, wage setting depends partly on beliefs about how the central bank responds to inflation. Changing the policy rule can change the parameters of an estimated behavioral relationship.\n\nA concrete example is a historical association between low unemployment and high inflation. If that association reflects occasional unexpected monetary expansions, it cannot necessarily be used to calculate how much permanently higher inflation would reduce unemployment. Once a higher-inflation policy becomes understood, wage demands and price-setting decisions adjust. Policy evaluation must account for this response rather than hold the historical relationship fixed.\n\nRational expectations nevertheless permits anticipated monetary policy to have real effects when prices or wages cannot adjust immediately. Stanley Fischer’s 1977 work on long-term contracts demonstrated the importance of this distinction. With contracts set at different dates, new policy information can affect real wages and employment even when agents form expectations rationally. In New Keynesian models, monetary policy influences spending partly through the expected real interest rate, while nominal rigidities prevent immediate adjustment to a purely nominal change.\n\nThis distinction also underlies forward guidance. An announcement about future interest rates can affect current borrowing costs and spending if it changes credible expectations of future policy. Its effect depends on what the announcement reveals, whether it is believed, how long people plan ahead, and the frictions in the economy. Rational expectations organizes those mechanisms; it does not establish that all announcements are credible or that every announced policy path is feasible."
    },
    {
      "heading": "Credibility, commitment, and policy regimes",
      "text": "Rational expectations shifted policy analysis from isolated interventions toward rules, institutions, and the credibility of future actions. If private decisions depend on expected policy, a government cannot evaluate a policy solely by its immediate mechanical effects. It must also account for how the public expects policy to respond after wages, prices, debts, or investments have already been chosen.\n\nFinn Kydland and Edward Prescott’s 1977 analysis of time inconsistency formalized a central difficulty. A policy plan that is desirable before private decisions are made may no longer be optimal for the policymaker afterward. For example, a monetary authority may announce low inflation to influence wage agreements but later prefer an unexpected expansion to stimulate output. If wage setters understand this incentive, the announcement alone does not secure low expected inflation.\n\nRobert Barro and David Gordon developed influential monetary-policy models of this problem in 1983. Under discretionary policymaking, an authority that seeks output above its sustainable level may generate an inflation bias: expected inflation is positive, but the attempted output gain disappears because it is anticipated. The result is conditional on the authority’s objectives and the model’s constraints. Rational expectations reveals the consequences of the incentive problem; it does not itself explain why the authority has those objectives.\n\nInstitutional responses include policy rules, reputation, delegation, and arrangements supporting central-bank independence. Inflation targeting can help organize expectations by providing a public objective and an accountability framework, although a numerical target alone does not ensure credibility. Credibility depends on observed behavior, institutional authority, and the perceived consistency of monetary policy with other government commitments.\n\nFiscal policy introduces additional interactions. The implications of a monetary announcement depend partly on expected government spending, taxation, and the financing of public debt. A promise to reduce money creation may be difficult to sustain if fiscal policy continually requires monetary financing. Conversely, a credible fiscal adjustment can alter expectations about future inflation and debt repayment before all of the adjustment occurs.\n\nRegime changes create a particularly demanding application of the hypothesis. In a conventional rational expectations model, agents may be assumed to know the new rule immediately. In practice, they may have to infer whether an announced change is permanent, whether institutions can implement it, and how policymakers will react to exceptional events. Models with uncertainty about policy types or regime transitions can represent this inference explicitly. Expectations can remain rational even when people initially assign only a limited probability to a new policy’s survival.\n\nCommitment and discretion also have different implications in forward-looking models. Under commitment, a policymaker can sometimes influence current behavior through promises about future actions that would be unattractive when that future arrives. Under discretion, private agents anticipate subsequent reoptimization. The difference is central to analyses of stabilization policy near the lower bound on interest rates, where expectations of future accommodation can affect current demand."
    },
    {
      "heading": "Financial prices, consumption, and intertemporal choice",
      "text": "Financial markets provide a natural setting for rational expectations because asset prices depend on uncertain future payoffs. A general pricing condition is\n\n\\[P_t=\\mathbb E_t\\!\\left[M_{t+1} \\left(D_{t+1}+P_{t+1}\\right)\\right],\\]\n\nwhere \\(P_t\\) is an asset’s price, \\(D_{t+1}\\) is its next-period payment, and \\(M_{t+1}\\) is a stochastic discount factor. The discount factor captures time preference and the valuation of payoffs in different states of the world. Rational expectations requires the conditional expectation to be consistent with the model’s process for payoffs, prices, and consumption.\n\nThis relationship connects rational expectations with the efficient-market hypothesis, but the concepts are not identical. Market efficiency concerns the incorporation of a specified information set into prices. Rational expectations concerns the consistency of beliefs with the economic process. Neither implies that observed prices must follow a simple random walk, because expected returns can vary with risk, preferences, and economic conditions.\n\nA risky asset can offer a predictably higher expected return without creating an arbitrage opportunity. Investors may require compensation for receiving low payoffs precisely when consumption is scarce. Consequently, predictable returns do not by themselves establish irrational beliefs. Likewise, a failure of a particular asset-pricing equation may reflect its model of risk rather than its expectations assumption. This is one reason empirical assessments often involve joint hypotheses.\n\nInformation acquisition creates another complication. Sanford Grossman and Joseph Stiglitz argued in 1980 that perfectly informationally efficient markets are problematic when gathering information is costly. If prices fully reveal all costly information, investors have little incentive to acquire it; without informed investors, prices cannot reveal it in the first place. Rational-expectations models with noisy supply or dispersed private signals can instead produce partially revealing prices, leaving an incentive to become informed. Information asymmetry is therefore compatible with rational expectations.\n\nThe hypothesis also transformed consumption theory. Under the permanent income hypothesis, households base spending on expected lifetime resources rather than current income alone. Robert Hall’s 1978 analysis showed that, under restrictive assumptions including quadratic utility and a suitable relationship between the interest rate and time preference, consumption follows a random walk. Information already available should not predict consumption changes because it should already have been incorporated into spending plans.\n\nThose restrictions matter. With more general preferences, uncertainty can affect precautionary saving, and consumption growth can depend on expected returns and risk. Borrowing constraints can prevent households from responding fully to future income, while durable goods and adjustment costs alter the timing of expenditure. Rational expectations supplies the consistency condition for forecasts within these models, not a universal prediction that consumption is unpredictable or that all households react identically to news."
    },
    {
      "heading": "Empirical tests and interpretation of forecast errors",
      "text": "The empirical appeal of rational expectations is that it restricts observable forecast errors. If a survey records a forecast \\(f_t\\) of a later outcome \\(x_{t+h}\\), a simple test examines\n\n\\[x_{t+h}=\\alpha+\\beta f_t+\\varepsilon_{t+h}.\\]\n\nUnder standard unbiasedness conditions, the null hypothesis is \\(\\alpha=0\\) and \\(\\beta=1\\). A stronger test asks whether variables known at the forecast date explain the error \\(x_{t+h}-f_t\\). Persistent predictability can be evidence against the joint claim that the forecast is a conditional mean and uses the specified information efficiently.\n\nSuch tests require careful matching of information and timing. A forecast prepared before an employment release cannot legitimately be evaluated as though the forecaster already knew that release. National accounting data are also revised, sometimes substantially. Researchers distinguish data available in real time from later revised estimates when testing forecasts of gross domestic product, inflation, and other variables. Otherwise, information unavailable to the forecaster can create a misleading appearance of irrationality.\n\nSurvey data provide more direct evidence about beliefs than do realized choices alone. Important sources include the Survey of Professional Forecasters, whose origins date to 1968, the University of Michigan Surveys of Consumers, and the European Central Bank’s Survey of Professional Forecasters, introduced in 1999. These surveys cover different populations, forecast horizons, and reporting formats. Professional forecasts cannot automatically be treated as measures of household beliefs, while an average across respondents need not correspond to any one respondent’s information set.\n\nForecast evaluation also depends on what respondents are reporting. A point forecast might represent a mean, median, most likely outcome, or a rounded answer. A person minimizing asymmetric costs of overprediction and underprediction may rationally report something other than the mean. Tests based on unbiased conditional-mean forecasts can then reject even when the answer is optimal under the respondent’s objective. Density forecasts, which assign probabilities to ranges of outcomes, provide additional evidence but introduce their own measurement challenges.\n\nMacroeconomic tests often assess rational expectations jointly with a structural model. Predictable errors in an estimated equation may reflect an omitted variable, a mistaken adjustment-cost specification, or an incorrect policy rule. A rejection does not uniquely identify expectations as the source of failure. Conversely, an inability to reject a restriction does not establish that agents know the full economic model.\n\nFinite samples create further difficulties. Forecasts for overlapping horizons have mechanically related errors, requiring appropriate statistical treatment. Rare disasters or policy transitions may be anticipated with small probability but not occur within the sample, producing apparent bias—the peso problem. Unexpected structural breaks can generate several errors in the same direction. Empirical work therefore examines not only average accuracy but also information use, revisions, disagreement, and the stability of forecast relationships across periods."
    },
    {
      "heading": "Learning, limited attention, and alternative expectations models",
      "text": "A persistent criticism of rational expectations concerns the knowledge it assigns to economic agents. In many implementations, people are treated as knowing the structure of the economy, its parameters, the policy regime, and the relevant probability distributions. Even when this is a useful equilibrium benchmark, it leaves open how such knowledge is acquired. Economists estimating the same relationships often disagree, revise their models, and face substantial parameter uncertainty.\n\nAdaptive learning addresses this issue by allowing agents to estimate forecasting rules from observed data and update them over time. For example, households may forecast inflation with a regression whose coefficients change as new observations arrive. Their decisions influence the data used in later estimation, creating feedback between beliefs and economic outcomes. Research associated with Albert Marcet, Thomas Sargent, George Evans, and Seppo Honkapohja examines conditions under which learning converges to a rational expectations equilibrium.\n\nConvergence is not automatic. An equilibrium can be mathematically well defined but unstable under a plausible learning rule. Conversely, sufficiently stable economic relationships may allow relatively simple forecasting procedures to approach model-consistent predictions. This distinction gives learning models a role in equilibrium selection and policy design: a rule that works under full knowledge may behave differently when agents are continually estimating how the economy operates.\n\nModels of limited information offer another approach. In sticky-information models, including work by N. Gregory Mankiw and Ricardo Reis in 2002, people update their information intermittently. Those who have not recently updated continue to act on older information, so the aggregate response to news is gradual. Christopher Sims’s work on rational inattention instead treats information processing as costly or capacity-constrained. Agents allocate attention across competing decisions rather than continuously processing every available signal.\n\nThese approaches need not abandon all forms of rational expectations. A person can form a model-consistent forecast conditional on a deliberately limited signal. What changes is the information structure and the explanation of why information remains limited. Rational inattention differs from merely assuming arbitrary forecasting errors: it seeks to derive selective attention from an optimization problem. Nevertheless, its aggregate predictions can differ substantially from those of a full-information model.\n\nBounded rationality and behavioral economics allow broader departures, including extrapolation from recent observations, overreaction to salient news, or persistent differences in perceived models. Diagnostic and experience-based expectations models investigate why particular events receive disproportionate weight. Experimental and survey research examines whether forecast revisions are too small or too large relative to the information that becomes available, and whether patterns vary across households, firms, and professional forecasters.\n\nThe practical modeling choice is therefore not simply between perfectly accurate forecasts and arbitrary mistakes. Alternatives differ in what agents know, how they acquire information, how they process it, and whether their forecasting rules eventually converge. Rational expectations remains a benchmark against which these mechanisms can be specified and tested, while models of learning and attention make the formation of expectations an explicit part of the economic explanation."
    }
  ],
  "links": [
    "macroeconomics",
    "inflation",
    "expected-value",
    "variance",
    "probability-distribution",
    "unemployment",
    "phillips-curve",
    "milton-friedman",
    "real-business-cycle-theory",
    "new-keynesian-economics",
    "nobel-memorial-prize-in-economic-sciences",
    "dynamic-stochastic-general-equilibrium",
    "monetary-policy",
    "lucas-critique",
    "central-bank",
    "real-interest-rate",
    "forward-guidance",
    "central-bank-independence",
    "inflation-targeting",
    "fiscal-policy",
    "taxation",
    "public-debt",
    "information-asymmetry",
    "saving",
    "gross-domestic-product",
    "bounded-rationality",
    "behavioral-economics"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}