{
  "id": "dynamic-stochastic-general-equilibrium",
  "title": "Dynamic Stochastic General Equilibrium",
  "lang": "en",
  "source_lang": "en",
  "summary": "Dynamic stochastic general equilibrium models explain aggregate economic fluctuations through interacting forward-looking agents, uncertainty, and economy-wide equilibrium conditions.",
  "category": "economics",
  "subcategory": "macroeconomics",
  "type": "concept",
  "aliases": [
    "DSGE",
    "DSGE modeling",
    "dynamic stochastic general equilibrium modeling"
  ],
  "generated_at": "2026-11-04",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/dynamic-stochastic-general-equilibrium/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/dynamic-stochastic-general-equilibrium/",
      "json": "https://aiwiki.page/zh/dynamic-stochastic-general-equilibrium.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Dynamic stochastic general equilibrium (DSGE) is a framework for constructing quantitative models of an economy in which households, firms, and public institutions interact over time under uncertainty. Such models combine behavioral equations, technological and institutional constraints, and conditions ensuring consistency across markets. They are widely used in macroeconomics to investigate economic fluctuations, interpret aggregate data, and compare the consequences of alternative policy arrangements.\n\nEach part of the name identifies a defining feature. Dynamic means that current decisions depend on inherited states and anticipated future conditions. Stochastic means that the economy is subject to uncertain disturbances. General equilibrium means that decisions in different markets are determined jointly rather than analyzed in isolation. A change in borrowing costs, for example, can affect consumption, investment, wages, prices, government revenue, and international transactions simultaneously.\n\nDSGE denotes a family of models rather than a single theory of economic behavior. Early versions emphasized technology and competitive markets; later versions incorporated sticky prices, financial constraints, unemployment, household heterogeneity, and departures from full-information expectations. Their distinguishing ambition is to connect aggregate outcomes to explicit decision problems while maintaining economy-wide consistency."
    },
    {
      "heading": "Intellectual origins and historical development",
      "text": "The intellectual foundations of DSGE modeling combine general equilibrium theory, intertemporal choice, stochastic processes, and quantitative analysis of the business cycle. Earlier equilibrium theory established how prices could coordinate decisions across markets. Growth theory supplied a framework for capital accumulation and long-run production, notably through the Solow–Swan model and optimal-growth models associated with Frank Ramsey, David Cass, and Tjalling Koopmans. Modern DSGE analysis brought these elements into economies repeatedly affected by uncertain events.\n\nA central methodological influence was the Lucas critique, articulated by Robert Lucas in 1976. Lucas argued that statistical relationships estimated under one policy regime need not remain stable when policy rules change, because households and firms alter their decisions and expectations. For example, a historical relationship between inflation and output cannot automatically be treated as a menu available to policymakers. DSGE models address this problem by specifying preferences, technologies, constraints, and policy rules separately. The intended distinction is between deeper behavioral parameters and equilibrium relationships that depend on the policy environment, although the stability of the supposedly deeper parameters remains an empirical question.\n\nAnother foundation was rational expectations, introduced by John Muth in 1961 and incorporated into macroeconomics during the 1970s. Under this assumption, agents’ forecasts are consistent with the model’s probability laws and available information. It does not imply perfect foresight: unexpected disturbances still produce forecast errors. It instead restricts systematic inconsistencies between the expectations assumed in individual decisions and the economy generated by those decisions.\n\nFinn Kydland and Edward Prescott’s 1982 article “Time to Build and Aggregate Fluctuations” became a landmark in real business cycle theory. Their model examined how technology disturbances, intertemporal substitution, and capital accumulation could generate fluctuations in a competitive economy. Its quantitative strategy emphasized choosing parameters from long-run evidence and microeconomic studies, then comparing simulated and observed business-cycle statistics. It established much of the computational architecture subsequently used in DSGE research.\n\nFrom the late 1980s onward, New Keynesian economics incorporated imperfect competition and nominal rigidities into related optimizing frameworks. Julio Rotemberg and Michael Woodford’s work in the 1990s, followed by estimated medium-scale models such as those developed by Lawrence Christiano, Martin Eichenbaum, and Charles Evans, and by Frank Smets and Rafael Wouters, helped establish the approach in monetary research. Smets and Wouters published influential euro-area and United States models in 2003 and 2007 respectively. The global financial crisis then intensified research on banking, financial amplification, household balance sheets, and nonlinear downturns."
    },
    {
      "heading": "The structure of a benchmark economy",
      "text": "A basic DSGE model starts with a household choosing consumption, work, and saving. A representative household might maximize expected lifetime utility,\n\n\\[E_0\\sum_{t=0}^{\\infty}\\beta^t \\left[ \\frac{C_t^{1-\\sigma}-1}{1-\\sigma} -\\chi\\frac{N_t^{1+\\varphi}}{1+\\varphi} \\right],\\]\n\nwhere \\(C_t\\) denotes consumption, \\(N_t\\) hours worked, \\(0<\\beta<1\\) the discount factor, and \\(\\sigma,\\chi,\\varphi\\) preference parameters. For \\(\\sigma=1\\), the consumption term is interpreted as \\(\\log C_t\\). This utility function captures a trade-off between consumption and leisure; other specifications introduce habit formation, durable goods, or preferences that separate risk aversion from intertemporal substitution.\n\nIn a closed economy with capital but no government, a simplified household budget constraint is\n\n\\[C_t+I_t=W_tN_t+R_t^kK_t+\\Pi_t, \\qquad K_{t+1}=(1-\\delta)K_t+I_t.\\]\n\nHere \\(I_t\\) is investment, \\(W_t\\) the real wage, \\(R_t^k\\) the rental return on capital, \\(\\Pi_t\\) distributed profits, and \\(\\delta\\) the depreciation rate. Capital available today is inherited from earlier investment. The household therefore cannot treat current and future consumption as independent decisions. Suitable solvency and transversality conditions exclude financing plans that rely indefinitely on unpaid obligations.\n\nA central optimality condition is the consumption Euler equation. If \\(R_t\\) is a gross real return known at date \\(t\\), and \\(u_C\\) denotes marginal utility of consumption, then\n\n\\[u_C(C_t)=\\beta E_t\\!\\left[R_tu_C(C_{t+1})\\right].\\]\n\nThe equation equates the utility cost of giving up one unit of consumption now to the expected discounted utility gained by investing it. With risky returns, the interaction between returns and future marginal utility matters: an asset paying especially well in adverse states can be valuable even if its average payoff is relatively low. A separate intratemporal condition connects labor supply to the real wage.\n\nFirms combine capital and labor through a production function, commonly\n\n\\[Y_t=A_tK_t^\\alpha N_t^{1-\\alpha},\\]\n\na Cobb–Douglas specification. The variable \\(A_t\\) represents total factor productivity. Under perfect competition, factor prices equal marginal products. With market power, pricing instead depends on markups over marginal cost. Investment adjustment costs, variable capital utilization, and sector-specific inputs can modify the production side.\n\nEquilibrium requires individual decisions to be mutually compatible. In the simplest closed economy, the goods-market condition is \\(Y_t=C_t+I_t\\); with government purchases it becomes \\(Y_t=C_t+I_t+G_t\\). Labor supplied must equal labor hired, and asset positions must satisfy their aggregate supply restrictions. An equilibrium is consequently a collection of decision rules, prices, allocations, and probability laws—not merely one price vector. The representative household is a tractability device, not a claim that actual households have identical incomes or circumstances."
    },
    {
      "heading": "Nominal rigidities and monetary transmission",
      "text": "A competitive model with fully flexible prices provides limited scope for explaining why changes in nominal policy rates affect real activity. New Keynesian DSGE models introduce monopolistic competition and constraints on price or wage adjustment. Firms produce differentiated goods and possess some market power, but cannot always reset their prices immediately. Changes in spending can therefore alter production and employment before all nominal prices adjust.\n\nOne widely used specification is Calvo pricing, named after Guillermo Calvo’s 1983 contribution. Each period, a firm has a constant probability \\(1-\\theta\\) of resetting its price, independent of the time since its last adjustment. The average duration of a price spell is \\(1/(1-\\theta)\\) periods. Thus, \\(\\theta=0.75\\) implies an expected duration of four quarters in a quarterly model. A resetting firm considers expected future demand and costs because its price may remain in force for several periods. An alternative, associated with Rotemberg, makes price changes costly rather than infrequent.\n\nNear a zero-inflation steady state, a basic version produces the New Keynesian Phillips curve:\n\n\\[\\pi_t=\\beta E_t\\pi_{t+1}+\\kappa x_t+u_t.\\]\n\nHere \\(\\pi_t\\) is inflation, \\(x_t\\) is the output gap, \\(\\kappa\\) measures the sensitivity of inflation to economic slack, and \\(u_t\\) is a cost-push disturbance. The direct theoretical relationship is often between inflation and real marginal cost; expressing it through the output gap requires additional assumptions. The relevant gap is usually output relative to a model-defined flexible-price allocation, not automatically output relative to a statistical trend.\n\nHousehold spending decisions yield a forward-looking demand equation,\n\n\\[x_t=E_tx_{t+1} -\\frac{1}{\\sigma} \\left(i_t-E_t\\pi_{t+1}-r_t^n\\right),\\]\n\nwhere \\(i_t\\) is the nominal policy rate and \\(r_t^n\\) the natural real interest rate consistent with the flexible-price allocation. A higher expected real interest rate, relative to that natural rate, encourages postponement of spending. The model therefore makes expectations of future policy important, rather than treating the current rate as the sole determinant of demand.\n\nA simple monetary rule completes the system:\n\n\\[i_t=\\rho i_{t-1} +(1-\\rho)(\\phi_\\pi\\pi_t+\\phi_xx_t) +\\varepsilon_t^i,\\]\n\nwith variables measured as deviations from their steady states. This is a smoothed version of a Taylor rule, named after John Taylor’s 1993 formulation. In standard settings, a sufficiently strong response to inflation helps produce a unique stable equilibrium, although the exact conditions depend on the full model and on fiscal arrangements. Wage stickiness, consumption habits, and investment adjustment costs are often added to reproduce the delayed responses observed in data. These additions make the monetary transmission mechanism richer than the three-equation benchmark."
    },
    {
      "heading": "Uncertainty, equilibrium computation, and dynamic responses",
      "text": "The stochastic component of a DSGE model specifies which disturbances occur and how they evolve. A common productivity process is\n\n\\[\\log A_t=(1-\\rho_A)\\log \\bar A +\\rho_A\\log A_{t-1}+\\varepsilon_t^A,\\]\n\nwhere \\(|\\rho_A|<1\\) and \\(\\varepsilon_t^A\\) is an innovation with zero conditional mean. Other processes represent government spending, investment efficiency, preferences, markups, foreign demand, or monetary policy. A disturbance’s economic interpretation must follow from the equation it enters. An unexplained residual in a pricing equation is not, by itself, direct evidence of a particular institutional change.\n\nThe model distinguishes predetermined state variables from forward-looking choices. Installed capital and outstanding debt are typical states; consumption and asset prices can adjust when information arrives. Equilibrium decision rules map the current state and new information into choices and next period’s state. This recursive representation makes it possible to simulate the economy for thousands of periods or calculate how a particular event propagates through time.\n\nMost medium-scale models are solved approximately. Researchers first calculate a deterministic steady state, or transform a growing economy into stationary variables around a balanced-growth path. They then approximate the equilibrium conditions locally. First-order linearization, often expressed in percentage deviations from steady state, converts a nonlinear system into a tractable linear rational-expectations problem. Olivier Blanchard and Charles Kahn’s 1980 analysis established influential conditions for the existence and uniqueness of solutions in such systems, relating unstable roots to the number of forward-looking variables.\n\nFirst-order approximation has an important limitation: under standard assumptions, it displays certainty equivalence. Conditional mean decision rules do not capture effects of shock variance on behavior, even though realized shocks generate fluctuations. Second- and higher-order methods can capture precautionary saving, risk-adjusted average outcomes, and some risk-premium effects. Projection methods approximate decision rules over a wider state space rather than only near one point. Models with occasionally binding borrowing constraints or interest-rate floors may require piecewise-linear or other nonlinear algorithms.\n\nAn impulse response function traces how model variables respond to a specified innovation. Its interpretation depends on the shock’s size, persistence, information content, and assumed future policy response. A one-time monetary tightening differs from news that rates will remain higher for several years. In nonlinear models, responses can also depend on initial conditions and shock direction. Forecast-error variance decompositions attribute fluctuations at particular horizons to modeled shocks; historical decompositions reconstruct past movements conditional on the estimated model and shock specification. These calculations describe a model’s internal account of the data, not independently verified causal histories."
    },
    {
      "heading": "Calibration, estimation, and empirical identification",
      "text": "DSGE models are connected to evidence through calibration, statistical estimation, or combinations of the two. Calibration assigns parameters using outside evidence or selected empirical targets. A quarterly discount factor might be chosen to imply a plausible average real return; depreciation can be informed by investment and capital-stock data. In a simple log-utility economy with a constant gross real return, \\(\\beta=0.99\\) implies a quarterly return near \\(1/\\beta-1\\), approximately 1 percent, or about 4.1 percent compounded annually. Growth, risk, taxes, and different preferences alter that calculation.\n\nCalibration was particularly prominent in early real business cycle research. Models were assessed using moments such as the variability of consumption relative to output, the cyclicality of hours, and correlations among investment, productivity, and production. These comparisons remain useful, but matching a few moments does not establish that the proposed mechanism is uniquely supported. Different mechanisms can generate similar aggregate statistics, while a model matching unconditional variation may fail to reproduce responses to a particular disturbance.\n\nLikelihood-based estimation connects the model to observed time series through a state-space representation. A linearized solution can be written as\n\n\\[s_t=Ts_{t-1}+R\\varepsilon_t, \\qquad y_t=d+Zs_t+\\eta_t,\\]\n\nwhere \\(s_t\\) contains latent states, \\(y_t\\) contains observed variables, and \\(\\eta_t\\) represents measurement error. The Kalman filter evaluates the likelihood in a linear Gaussian setting while updating estimates of unobserved states. Nonlinear or non-Gaussian models may require other filtering methods. The measurement equations matter: a model variable for hours per person is not interchangeable with total employment, and model consumption need not correspond to every component in published expenditure data.\n\nBayesian inference combines this likelihood with prior distributions over parameters. Posterior simulation is widely used because DSGE likelihoods can be complicated and the parameter space large. Priors can incorporate microeconomic evidence or restrict implausible regions, but posterior conclusions may be sensitive to them when the aggregate data contain little information about a parameter. Posterior credible intervals describe uncertainty conditional on the model, prior, and data treatment.\n\nIdentification concerns whether distinct parameter values or mechanisms imply distinguishable observations. Price stickiness, wage stickiness, habits, and adjustment costs may produce overlapping patterns, making their individual contributions difficult to estimate precisely. Stochastic singularity presents another problem: a model with too few independent innovations can imply exact relationships among observed series that the data violate. Additional economically motivated disturbances, measurement errors, or a smaller observable set can address the issue, but each changes interpretation. Empirical evaluation therefore includes out-of-sample forecasting, sensitivity analysis, comparisons with less restrictive statistical models, and checks against household- or firm-level evidence."
    },
    {
      "heading": "Policy analysis and institutional applications",
      "text": "DSGE models are particularly suited to conditional questions: how would inflation and output evolve if a specified disturbance occurred, if a policy rule changed, or if an announced measure were temporary rather than permanent? Because households and firms respond to anticipated future conditions, policy experiments must specify both the immediate intervention and its expected continuation. An unanticipated one-quarter tax change is not equivalent to a permanent reform announced several years in advance.\n\nIn monetary policy analysis, models compare interest-rate rules, inflation targets, communication strategies, and stabilization trade-offs. Forward guidance operates through expected future real rates and their influence on present spending. Some representative-agent models generate very large effects from promises concerning distant future rates, a result often called the forward-guidance puzzle. Research on finite planning horizons, uncertainty about future policy, incomplete information, and household borrowing constraints examines why those effects may be smaller or transmitted differently.\n\nFor fiscal policy, the government’s intertemporal budget constraint is essential. Higher public purchases must eventually be associated with taxes, lower spending elsewhere, debt adjustment, or other financing consistent with the model. The fiscal multiplier is therefore not a universal parameter: it varies with the instrument, financing, monetary response, economic slack, and household access to credit. Ricardian equivalence can emerge under restrictive conditions, including sufficiently forward-looking households and suitable tax arrangements. Borrowing constraints, finite lives, distortionary taxation, and redistribution can prevent that result.\n\nWelfare analysis evaluates policies using the preferences represented in the model. In selected New Keynesian environments, a second-order approximation produces a loss function involving inflation variation and deviations of output from an appropriate benchmark. But the weights and even the relevant stabilization targets depend on distortions and assumptions. An output gap relative to flexible-price production need not equal the gap relative to the socially efficient allocation. Where households differ, welfare comparisons also depend on who receives income, pays taxes, or bears inflation and employment risk, linking policy evaluation to welfare economics.\n\nCentral banks and international institutions have developed DSGE models as components of broader analytical systems. Examples include the European Central Bank’s New Area-Wide Model, the Bank of England’s COMPASS framework, and the International Monetary Fund’s Global Integrated Monetary and Fiscal Model. Institutional use does not imply exclusive reliance on DSGE forecasts: such models are commonly combined with time-series tools, sectoral analysis, survey information, and expert judgment. Their distinctive role is to organize assumptions and produce internally consistent scenarios, while exposing how results depend on behavioral equations and policy arrangements."
    },
    {
      "heading": "Financial frictions, heterogeneous agents, and international linkages",
      "text": "Financial extensions depart from the frictionless benchmark in which financing arrangements have little independent influence on real allocation. With asymmetric information, limited enforcement, or costly monitoring, borrowers may face an external-finance premium that depends on their net worth. In the financial accelerator mechanism associated with Ben Bernanke, Mark Gertler, and Simon Gilchrist, a negative shock reduces borrower net worth, raises financing costs, lowers investment, and further weakens balance sheets. Their influential quantitative treatment appeared in 1999.\n\nCollateral constraints provide a related but distinct channel. In Nobuhiro Kiyotaki and John Moore’s 1997 credit-cycle model, assets can serve both as productive inputs and as collateral. Falling asset prices tighten borrowing limits, forcing expenditure reductions or asset sales that can depress prices further. DSGE models with banks additionally distinguish deposit funding, bank net worth, loan pricing, and capital requirements. Such distinctions are important for studying macroprudential policy, although a model must explicitly represent relevant institutions before its results can be interpreted as evidence about them.\n\nHousehold heterogeneity changes both aggregate dynamics and distributional analysis. In representative-agent models, a single household summarizes consumption and saving decisions. In heterogeneous-agent models, households differ in earnings, wealth, employment status, or asset access. They may face uninsurable income risk and borrowing limits, building precautionary savings when possible. Per Krusell and Anthony Smith’s 1998 work demonstrated computational methods for economies with both aggregate uncertainty and distributions of individual wealth.\n\nHeterogeneous-agent New Keynesian models, often abbreviated HANK, combine these household distributions with nominal rigidities. Monetary policy then affects consumption not only through intertemporal substitution but also through employment income, interest receipts and payments, and changes in asset values. A household with little liquid wealth can react strongly to a temporary income gain even if it owns illiquid assets. Consequently, the distribution of the marginal propensity to consume becomes central to transmission. The distribution itself evolves over time and is part of the aggregate state, substantially increasing computational demands.\n\nOpen-economy DSGE models add imported consumption and inputs, foreign assets, international risk sharing, and exchange-rate determination. Policy effects depend on the currency in which firms set prices, the degree of exchange-rate pass-through, and the availability of external finance. In a currency union, member economies share monetary policy while potentially retaining different fiscal systems, financial conditions, and production structures. Multisector models introduce input-output linkages and sector-specific pricing; climate-oriented models can connect emissions, energy use, and investment. In each extension, the relevant mechanism comes from the added constraints and market structure rather than from the DSGE label alone."
    },
    {
      "heading": "Interpretation, limitations, and continuing research",
      "text": "DSGE modeling offers accounting consistency and an explicit mapping from assumptions to outcomes, but these features do not guarantee an empirically adequate explanation. Optimization can be based on unrealistic information or decision-making assumptions; market-clearing equations can coexist with unemployment or rationing only if the model explicitly represents the relevant mechanisms. A formally coherent model may omit the institution or behavior most important for the question under investigation.\n\nExpectations are a major area of debate. Rational expectations provides a disciplined equilibrium benchmark, but actual households and firms may have incomplete information, disagreement, limited attention, or imperfect knowledge of the economy’s structure. Models incorporating bounded rationality, adaptive learning, noisy signals, or survey-based expectations investigate these departures. Such alternatives require their own restrictions: replacing rational expectations with unrestricted forecasting behavior can make the framework difficult to identify or evaluate. Survey evidence can help distinguish beliefs about inflation, income, and policy from the expectations implied by a fitted model.\n\nAggregate shock identification is another persistent limitation. A model may attribute an economic contraction to a “risk-premium shock,” for example, without explaining the institutional events behind that disturbance. Highly flexible shock processes can improve fit while weakening the explanatory role of the underlying mechanisms. External instruments, natural experiments, administrative data, and microeconomic estimates can provide additional restrictions, although transporting estimates from one population or policy environment to another requires care. Historical decompositions are especially sensitive to the set of disturbances allowed in the model.\n\nLarge disruptions expose the limitations of local approximations and stable-regime assumptions. Financial crises can involve default, asset fire sales, institutional failure, and binding funding constraints; abrupt shutdowns can impose physical restrictions on production and contact-intensive consumption. A linearized model estimated around ordinary conditions may not represent these states adequately. Nonlinear methods, explicit default mechanisms, sectoral detail, and regime-switching specifications can address particular deficiencies, but introduce additional computational and identification challenges. Greater complexity does not necessarily improve forecast accuracy or make competing mechanisms easier to distinguish.\n\nEvaluation consequently depends on the purpose of the model. A small model may clarify a theoretical transmission channel, while a larger estimated model may be designed for forecasting or institutional scenario analysis. Policy conclusions require attention to parameter uncertainty, uncertainty about future shocks, and uncertainty about the model’s structure. Comparisons across alternative specifications can reveal which results depend on a particular friction or expectations assumption. Continuing research links aggregate models to household and firm data, develops faster methods for heterogeneous-agent economies, and investigates how expectations, production networks, and financial constraints interact outside the neighborhood of a steady state."
    }
  ],
  "links": [
    "macroeconomics",
    "general-equilibrium-theory",
    "business-cycle",
    "lucas-critique",
    "rational-expectations",
    "real-business-cycle-theory",
    "new-keynesian-economics",
    "global-financial-crisis",
    "utility-function",
    "budget-constraint",
    "marginal-utility",
    "production-function",
    "cobb-douglas-production-function",
    "total-factor-productivity",
    "monopolistic-competition",
    "market-power",
    "phillips-curve",
    "marginal-cost",
    "real-interest-rate",
    "taylor-rule",
    "impulse-response",
    "calibration",
    "kalman-filter",
    "bayesian-inference",
    "monetary-policy",
    "forward-guidance",
    "fiscal-policy",
    "fiscal-multiplier",
    "welfare-economics",
    "information-asymmetry",
    "net-worth",
    "collateral",
    "macroprudential-policy",
    "marginal-propensity-to-consume",
    "exchange-rate",
    "bounded-rationality"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}