Baseline Macro ABM (K+S Mark I) Model

The K+S (Keynes meets Schumpeter) Mark I model, introduced by Dosi, Fagiolo, and Roventini (2010), is the baseline closed-economy macro ABM. It was built to answer a question that neither RBC nor New Keynesian DSGE models handle well: can aggregate fluctuations, fat-tailed growth-rate distributions, and power-law firm sizes all emerge from the decentralized interaction of boundedly rational firms without imposing any exogenous shock process or representative-agent aggregation? The answer is yes, and the mechanism runs through Schumpeterian innovation at the micro level feeding into Keynesian demand multipliers at the macro level. The model descends from the evolutionary economics tradition of Nelson and Winter (1982) and the Keynesian-structuralist lineage of Dosi (1988), but it was the first to weld both into a single closed macro system with explicit labor, credit, and goods markets. Key predecessors include Russo et al. (2007) and the Eurace project; key successors include the K+S Mark II (Dosi et al. 2013), which adds fiscal and monetary policy, and the Dawid-Delli Gatti (2018) survey that positions K+S Mark I as the reference macro ABM.

The core mechanism has two interlocking loops. Loop 1 is Schumpeterian: capital-goods firms invest in R&D, generate stochastic innovations that improve machine productivity, and sell new-vintage machines to consumption-goods firms. A firm that draws a good innovation captures market share through lower unit costs, grows, hires more workers, and generates demand for other firms' output. Loop 2 is Keynesian: aggregate demand is the sum of wage income (workers spend all wages) and firm investment. When a cluster of consumption-goods firms faces weak demand, they cut production, fire workers, reduce investment orders from the capital-goods sector, and the contraction propagates through the income-expenditure multiplier. Business cycles emerge from the interaction of these two loops: innovation clusters create booms, demand shortfalls create busts, and the amplitude depends on the distribution of firm productivities and financial positions rather than on any imposed stochastic forcing.

The model has been used across several institutional settings. Dosi et al. (2010, 2013, 2015, 2017) at the Sant'Anna School of Advanced Studies have produced the core sequence of K+S papers. The European Commission's Joint Research Centre has used K+S-family models for industrial policy evaluation. The OECD has referenced ABM-generated results on inequality and growth in its policy reports. The Bank of Italy has explored K+S-type structures for financial stability analysis. Academic adoption spans computational economics (Journal of Economic Dynamics and Control, Journal of Economic Behavior and Organization, Journal of Evolutionary Economics) and policy-oriented outlets (Industrial and Corporate Change, Research Policy). The model's role in the ABM ecosystem is analogous to the Solow model in neoclassical growth or the three-equation NK model in DSGE: it is the simplest structure that closes all markets and generates the canonical stylized facts, and nearly every subsequent macro ABM is either an extension of it or a reaction against it.

The model family has evolved along several axes since the 2010 baseline. Mark II (Dosi et al. 2013) adds a government sector with fiscal policy, a central bank with a Taylor-type interest rate rule, and unemployment benefits. Mark III variants (Dosi et al. 2015, 2017) introduce an explicit banking sector with endogenous credit rationing and capital adequacy constraints, allowing the model to address financial crises and macroprudential policy. Further extensions include labor market institutions (minimum wages, collective bargaining), international trade (two-country K+S), climate damage functions, and skill-biased technical change. The active research frontier includes machine-learning-assisted calibration (Lamperti et al. 2018), surrogate-model estimation, and coupling with input-output networks.

The economy consists of three agent populations plus market mechanisms. Population 1 is capital-goods firms (F1 firms, typically 50 in research runs): each carries a technology frontier (the best machine productivity it can produce), an R&D budget rule, a pricing rule, and a client list of consumption-goods firms. Population 2 is consumption-goods firms (F2 firms, typically 200): each holds an inventory of machines with heterogeneous vintages and productivities, sets output based on expected demand, hires labor to operate machines, orders new machines from capital-goods firms to expand or replace capacity, and finances investment from internal funds or bank credit. Population 3 is workers/consumers (N, typically 1,000 to 5,000): each supplies one unit of labor inelastically, earns a wage, and consumes all income. A single bank provides credit to consumption-goods firms subject to a credit multiplier constraint.

Interaction occurs through three markets cleared each period. The capital-goods market is a customer-market mechanism: F2 firms send orders to their incumbent F1 suppliers, and occasionally switch suppliers if a competing F1 firm offers a lower price or better machine productivity. The consumption-goods market clears through a competitiveness-based replicator dynamic: each F2 firm's market share evolves as a function of its price and unfilled-demand ratio relative to the industry average. Firms with lower prices and fewer stockouts gain share; firms with higher prices or frequent stockouts lose share. The labor market is a sequential-hiring mechanism: firms post vacancies proportional to desired production, workers are matched to vacancies (with some friction), and the aggregate wage adjusts based on labor market tightness and productivity growth.

State variables update each period in a fixed sequence: (1) F1 firms invest in R&D and draw innovation outcomes, (2) F1 firms set prices for new machines, (3) F2 firms form demand expectations and plan production, (4) F2 firms compute investment demand (expansion + replacement), (5) F2 firms apply for bank credit if internal funds are insufficient, (6) the bank screens loan applications, (7) F2 firms order machines from F1 suppliers, (8) the labor market clears, (9) F2 firms produce and sell, (10) market shares update via the replicator dynamic, (11) wages update, (12) profits are computed and retained earnings accumulated, (13) entry and exit are processed. This 13-step sequence is not arbitrary: the ordering ensures that investment decisions are conditioned on current-period financial positions and that production is conditioned on actual labor availability.

The primary use of the K+S Mark I and its descendants is as a laboratory for macroeconomic policy experiments where heterogeneity and nonlinearity matter. Dosi et al. (2013) used the Mark II extension to compare fiscal austerity versus fiscal stimulus in a recession, finding that austerity deepens and lengthens downturns because the demand multiplier operates through the firm-size distribution: cutting government spending disproportionately hits small, credit-constrained firms that are already at the margin of exit, triggering a cascade of firm deaths and unemployment that is invisible in representative-agent models. The European Commission's JRC has used K+S-family models to evaluate industrial policy instruments (R&D subsidies, patent policy, competition policy) in settings where the distribution of firm capabilities is the policy-relevant object rather than the representative firm's optimality condition.

The model also works as a benchmark for validating ABM methodology itself. Lamperti et al. (2018) used K+S Mark I as the test case for machine-learning-assisted calibration (training a surrogate model on simulation output to speed up estimation). Barde and Van der Hoog (2017) used it to compare different approaches to ABM validation (simulated method of moments, Bayesian estimation, indirect inference). Dosi et al. (2018) used it to study which micro-level behavioral rules are necessary and sufficient to reproduce the canonical macro stylized facts (Okun's law, Beveridge curve, firm-size distribution, growth-rate distribution). In this methodological role, K+S Mark I functions like the Solow model does for neoclassical growth theory: the simplest structure you need to understand before building anything more complicated.

The model is not appropriate for several classes of questions. It cannot address asset pricing, financial market microstructure, or housing dynamics because it has no financial asset market. It cannot address monetary policy transmission because the baseline has no interest rate channel (the Mark II extension adds a rudimentary Taylor rule, but it is still thin compared to a full NK DSGE). It cannot address supply-chain disruptions because the two-sector structure has no input-output network. It is poorly suited to short-run forecasting because ABMs are not optimized for point prediction - they are scenario generators. For questions about optimal monetary policy, use a NK DSGE. For questions about financial contagion, use a firm-bank ABM or a network model. For forecasting, use a BVAR or a dynamic factor model.