Jan Tinbergen built the first macroeconometric model for the Netherlands in 1936, then extended the approach to the US economy for the League of Nations in 1939. Lawrence Klein refined the methodology at the Cowles Commission in 1944-47, producing Klein Model I; the Klein-Goldberger model (1955), built at the University of Michigan, extended it into a system of roughly 20 simultaneous equations estimated by limited-information maximum likelihood. Klein won the 1980 Nobel Prize for this work. The approach matured into the large-scale models of the 1960s-70s: the Wharton model (Klein and Evans), the Brookings model, the MPS model at the Federal Reserve, and Ray Fair's multicountry system. These models contained 50-500 equations covering consumption, investment (by sector), housing, international trade, government budgets, wage-price dynamics, and financial markets.
The core mechanism: each behavioral equation represents one economic relationship - a consumption function, an investment accelerator, a Phillips-curve wage equation, an import demand function. The equations are simultaneous because endogenous variables appear on the right-hand side of other equations. GDP is determined by the national accounts identity (C + I + G + NX), but C depends on GDP through the consumption function, and I depends on GDP through the accelerator. This circularity makes single-equation OLS biased. The Cowles Commission's solution was simultaneous-equations estimation: two-stage least squares (2SLS), three-stage least squares (3SLS), or full-information maximum likelihood (FIML).
The Lucas critique (1976) challenged the entire enterprise. If behavioral parameters shift when policy changes, a model estimated under one regime cannot predict outcomes under another. Sims (1980) attacked from a different angle: the identifying restrictions in macroeconometric models are 'incredible' - the choice of which variables to exclude from which equations is arbitrary and drives the results. These critiques triggered the profession's pivot toward VARs and DSGE. But large-scale models never disappeared from institutional forecasting. The Federal Reserve's FRB/US, NIESR's NiGEM (NIESR is an independent institute, not the Bank of England, whose in-house models were BEQM and then COMPASS), and Moody's Analytics macro model are all descendants of the Cowles tradition, updated with rational expectations, error correction, and flexible functional forms. The European Commission's QUEST III, by contrast, is an estimated open-economy DSGE (Ratto, Roeger, and in 't Veld 2009) - the micro-founded successor to the Commission's earlier QUEST systems.
Modern macroeconometric systems differ from their 1970s ancestors in three ways. First, expectations formation is more sophisticated: some equations use model-consistent forward-looking expectations, others use VAR-based or survey-based expectations. Second, long-run properties are anchored by error-correction mechanisms or calibrated steady states, preventing the model from drifting at long horizons. Third, stochastic simulation replaces deterministic baseline-plus-multiplier analysis. The model is shocked with draws from the estimated residual distribution to produce fan charts and probability assessments.