Eurace Large-Scale Macro ABM Model

Most macro models simplify the economy into one or two markets and ask what happens when you perturb them. The Eurace project asked: what happens when you build the core markets - goods, labor, capital goods, credit, and finance - from the bottom up, populate them with thousands of heterogeneous agents, and let the macro dynamics emerge? The project ran as an FP6 STREP grant from 2006 to 2009 (not, as is sometimes written, across FP6 and FP7), with the original architecture laid out in Deissenberg, van der Hoog and Dawid (2008). The name is not a portmanteau of 'European' and 'race': EURACE is Europe plus ACE, where ACE stands for Agent-based Computational Economics, the research programme the project belongs to. The model was developed across multiple institutions: Bielefeld (Dawid), Genoa (Cincotti), and the Catholic University of Milan (Delli Gatti), each contributing modules that were integrated into a single simulation platform.

The core mechanism is decentralized interaction across interconnected markets with no Walrasian auctioneer. Firms produce goods using labor hired in a search-and-matching labor market and capital bought from capital-goods producers, finance investment through retained earnings or bank credit, and sell output in a goods market where prices are posted and revised adaptively. Capital-goods producers build machines with labor and sell them to consumption-goods firms, which is what gives investment spending a counterparty on the selling side. Banks collect deposits from households, extend credit to firms subject to capital adequacy constraints, hold government bonds, and trade reserves in an interbank market. A central bank sets the policy rate and provides liquidity. A government collects taxes, pays unemployment benefits, and funds any deficit by issuing bonds that banks and households buy. Households earn wages, receive dividends and transfers, consume goods, save in bank deposits (and, in the original EURACE / Genoa branch, in firm equity), and search for jobs when unemployed. Each market clears through its own mechanism - posted prices in goods, search in labor, credit rationing in lending - and the aggregate business cycle emerges from their joint operation without any representative-agent shortcut. In the original EURACE and its Eurace Financial descendant there is in addition an order-book stock exchange where households trade firm equity; the Eurace@Unibi branch has no such exchange, and results from one branch should not be quoted as properties of the other.

Eurace has been used for policy analysis by academic research groups across Europe, and the project itself was funded by the European Commission - which is a different thing from the Commission having adopted the model as an operational tool, a distinction that is often blurred when the model is described. Dawid, Harting and Neugart (2014) study economic convergence and cohesion across regions; Dawid, Harting and Neugart (2018) study cohesion policy and its distributional consequences, including inequality. Cincotti, Raberto and Teglio (2010, 2012) developed the financial market module and studied how equity market volatility feeds back into firm investment through balance-sheet effects; Teglio, Raberto and Cincotti (2012) is their documented regulation work, on Basel-style bank capital requirements. The model has also been applied to labor market reforms and to monetary policy transmission. Dawid et al. (2019) is the transparency-and-reproducibility paper for the Eurace@Unibi model, not an innovation-policy study. It sits alongside the CATS/K+S model of Dosi et al. as one of the two canonical large-scale macro ABMs in the European tradition.

The model has evolved through several branches, and the distinction is load-bearing rather than bibliographic. Eurace@Unibi (the Bielefeld branch) emphasizes labor market dynamics and regional heterogeneity, with firms and workers located in a small number of regions and matching constrained by region; it has a capital-goods sector and a banking sector but NO order-book equity exchange. Eurace Financial (the Genoa branch), descending most directly from the original EURACE, emphasizes the financial market and banking sector, with a full order-book stock exchange and interbank lending. Neither published baseline contains a housing market: housing appears only in later variant work and should be read as an extension, not as one of the model's core markets. More recent work has added pension systems and climate-economy interactions. The active frontier includes Eurace extensions for studying green transition policies, pandemic shocks, and cross-country trade in a multi-region setup.

The economy contains six agent populations. Households (typically 1,600 to 6,400 in standard configurations, scalable to 100,000+) carry state vectors including: income, wealth decomposed into bank deposits, government bonds and (in the order-book branch) equity holdings, employment status, reservation wage, consumption budget, and a search radius for labor and goods markets. Consumption-goods firms (typically 80 to 320) carry: production technology, capital stock, inventory, labor force, wage bill, bank loans outstanding, retained earnings, posted price, and an adaptive pricing/production rule. Capital-goods producers (a small population, typically 1 to 10) carry: technology, labor force, posted machine price, and an order book of firm investment demand - they are the counterparty that firms actually buy capital FROM, and without them investment spending would leave one side of the transaction empty and break the model's own stock-flow-consistency claim. Banks (typically 2 to 20) carry: deposit base, loan portfolio, government bond holdings, capital ratio, interbank position, and credit rationing rule. A single central bank sets the policy rate via a Taylor-type rule and provides standing facilities. A single government collects income and corporate taxes, pays unemployment benefits, may run counter-cyclical fiscal policy, and funds any deficit by ISSUING BONDS that banks and households purchase - government debt has a holder, not just an outstanding balance.

Markets operate through distinct clearing mechanisms, each running on its own schedule within the period. The goods market uses a decentralized posted-price mechanism: firms set prices based on inventory levels and past sales, households visit a limited number of firms (search friction), and transactions clear bilaterally. The labor market uses search-and-matching: unemployed workers send applications to a limited number of firms, firms rank applicants and make offers, workers accept the best offer above their reservation wage. The capital-goods market is an order market: consumption-goods firms send investment orders to capital-goods producers at the posted machine price, and producers hire labor to fill them. The credit market uses credit rationing: firms apply to banks for loans, banks evaluate the firm's financial position against a capital-adequacy-constrained lending rule, and grant a fully or partially approved amount. The bond market is an absorption rule: banks and households take up new government issuance at the prevailing rate. The interbank market uses bilateral lending between banks with excess and deficit reserves. In the original EURACE / Eurace Financial branch there is additionally a financial market running a continuous double auction with limit and market orders; that market is absent from Eurace@Unibi.

The period structure proceeds in a fixed sequence each month (the canonical time step). Within each month: (1) firms set production plans based on expected demand net of inventory, (2) firms post vacancies and the labor market clears via search-and-matching, (3) firms produce using hired labor and installed capital, (4) firms set posted prices and the goods market clears, (5) firms compute investment demand and place orders in the capital-goods market; capital-goods producers hire and deliver, (6) firms apply for credit if cash flow is insufficient, the credit market clears, (7) defaults are processed for firms and banks, (8) the government collects taxes, pays transfers, and issues bonds to cover the deficit; banks and households absorb the issuance, (9) in the order-book branch only, the financial market runs for a configurable number of trading rounds, (10) the central bank updates the policy rate from year-over-year inflation, (11) aggregate statistics are recorded. This sequential structure makes the propagation path from any shock fully traceable, and every step names both sides of each transaction, which is what makes the stock-flow-consistency claim checkable rather than decorative.

The distributional composition of fiscal policy is the model's best-known application. The mechanism is robust and easy to state: a transfer directed at low-income households raises aggregate demand more than the same transfer spread uniformly, because low-income agents have higher marginal propensities to consume and spend at capacity-constrained firms, triggering hiring and investment cascades. That mechanism is invisible in representative-agent models where all households share one MPC. Two claims that usually travel with it do not belong: the specific figure that the targeted multiplier is '1.5x larger' is not traceable to a published Eurace result and is not quoted here, and the European Commission FUNDED the Eurace project rather than USING it operationally - funding and adoption are different facts and conflating them overstates the model's institutional standing. On the regional-policy side, Dawid, Harting and Neugart (2014) is the paper on economic convergence and cohesion across regions, and Dawid, Harting and Neugart (2018) is the paper on cohesion policy and inequality; results built on one should not be attributed to the other.

Cincotti, Raberto, and Teglio used the Eurace Financial variant to study the interaction between financial market volatility and real investment. When equity prices drop sharply, firm market capitalization falls, bank assessments of firm creditworthiness deteriorate (mark-to-market collateral effects), and credit tightening feeds back into reduced investment and lower output. Their documented regulation work is on bank capital requirements: Teglio, Raberto and Cincotti (2012) study how Basel-style capital ratios shape credit supply, leverage and the amplitude of the cycle, finding that tighter requirements trade lower crisis frequency against slower expansions. Financial-transaction-tax results are sometimes attributed to this group - a small Tobin tax reducing volatility without dampening investment, larger ones impairing price discovery - but no such Eurace study is traceable in their published record, and the claim is not repeated here.

The model has clear limitations. First, computational cost: a single run of the full Eurace model with 6,400 households, 320 firms, and 20 banks takes 5 to 30 minutes depending on the number of periods and financial market trading rounds. Full estimation (SMM or ABC) with 10,000+ evaluations takes weeks on a cluster. Second, calibration complexity: the model has 50 to 100 free parameters spread across its several markets, and the identification surface is severely multimodal. Most published Eurace work uses calibration to stylized facts rather than formal estimation. Third, the model does not handle expectation formation at the sophistication level of DSGE models - there is no forward-looking optimization, so policy announcements and credibility effects are absent. For questions where rational expectations and forward-looking behavior are essential (e.g., optimal monetary policy under commitment), a DSGE model is the right tool.