K+S Schumpeter-Keynes Growth ABM Model
How do firm-level R&D investment, Schumpeterian innovation cascades, and Keynesian aggregate demand failures jointly produce the statistical properties of long-run growth - trend growth, business cycles, fat-tailed recessions, and persistent productivity dispersion - without imposing them from above?
Understand the mathematics
Follow the same equations and assumptions one section at a time or in a continuous reading.
Hover or focus a symbol defined in this model to find its exact matches. Select it to keep the highlight; Escape clears the selection.
Odd+D Protocol: Model Specification
The ODD+D (Overview, Design concepts, Details + Decision-making) protocol documents the K+S model following the standard introduced by Grimm et al. (2006, 2010) and extended by Muller et al. (2013). The extended K+S specification as described across Dosi et al. (2010, 2013, 2015, 2017, 2018).
PURPOSE: The model explores how endogenous technological change (Schumpeterian innovation and imitation at the firm level) interacts with Keynesian aggregate demand dynamics (income-expenditure feedback, credit constraints, unemployment) to produce emergent macroeconomic growth patterns - including trend growth rates, business cycle fluctuations, fat-tailed recession distributions, and persistent cross-firm productivity dispersion - that cannot be derived from a representative-agent growth model.
ENTITIES, STATE VARIABLES, AND SCALES: The model contains five entity types. Capital-goods firms ( in calibrated runs) carry state variables: R&D expenditure, innovation probability, imitation probability, machine catalog (list of vintages with associated productivities and prices), revenue, and profit. Consumption-goods firms () carry: capital stock (portfolio of machine vintages), desired output, actual output, labor demand, employment, inventory, price, markup, market share, competitiveness score, debt, and equity. Workers ( to ) carry: employment status, wage, and consumption. Banks ( in baseline, multiple in extensions) carry: loan portfolio, capital ratio, and lending threshold. The government carries: tax revenue, spending, and debt. Temporal scale is quarterly periods over 200 to 600 quarters; spatial structure is absent in the baseline.
PROCESS OVERVIEW AND SCHEDULING: Each period proceeds in fixed order: (1) Capital-goods firms invest in R&D, draw innovation and imitation outcomes, and update their machine catalogs. (2) Consumption-goods firms form demand expectations (adaptive), determine desired production, compute labor demand and investment demand. (3) Consumption-goods firms place machine orders with capital-goods firms (choosing the best price-productivity ratio). (4) Credit market clears: firms with insufficient internal funds apply for bank loans; banks screen applications by leverage. (5) Labor market clears: firms hire or fire to match production plans; wages adjust to labor market tightness and inflation. (6) Production occurs. (7) Goods market clears: output is sold based on market shares; unsold output becomes inventory. (8) Market shares update via the replicator dynamic based on competitiveness. (9) Profits are computed, loans serviced, taxes paid, government transfers distributed. (10) Exit and entry: firms with persistently negative equity exit and are replaced by entrants. (11) Aggregate statistics are recorded.
Innovation productivity draw. Conditional on a successful R&D search, the candidate vintage has productivity equal to the firm's current best times one plus the increment. The increment comes from a Beta variate shifted and scaled onto the two-sided support , so a search can produce a worse technique as well as a better one. The support is two-sided on purpose: it is what makes the over innovation draw, imitation draw and incumbent technology a real operation rather than a formality.
Innovation and imitation probabilities. The R&D budget is split once: a share funds the innovation search and the remaining funds the imitation search, so the same money is never spent twice. Both hazards are increasing and concave in their own budget, reflecting diminishing returns to search effort. Imitation is typically cheaper per unit of hazard (higher ) but bounded by the existing frontier.
Market share replicator. The benchmark is the SHARE-WEIGHTED average competitiveness, not the unweighted mean: with an unweighted mean the shares drift off the simplex and the goods-market accounting breaks. Because is a sum of penalties it is negative by construction, so the deviation is divided by ; dividing by itself would flip the sign and shrink the share of every above-average firm. The explicit renormalization in the last step is required, not decorative - the growth factors do not sum to one on their own. The speed controls selection intensity.
Nelson-Winter evolutionary foundations
The K+S model is a direct descendant of Nelson and Winter (1982). The core architecture - firms following routines, stochastic innovation search, market selection - comes directly from their evolutionary theory of economic change. What K+S adds is the demand side: Nelson-Winter models are pure supply-side evolutionary models where demand is always sufficient. The K+S innovation was to show that adding Keynesian demand failures to the evolutionary framework changes growth dynamics qualitatively and quantitatively.
Why Beta for innovation increments
The Beta distribution for innovation step sizes is chosen because it is bounded: after shifting and scaling it lives entirely on , so no single draw can move the frontier by an arbitrary amount, and a failed search is a bounded setback rather than a catastrophe. It is also flexible in shape. For , puts mass near the UPPER end of the unit interval and puts mass near the lower end, so the 'mostly small increments with occasional larger ones' profile requires - for example . Setting would do the opposite and make large steps the norm. An unbounded distribution such as a Gaussian would occasionally produce implausibly large jumps in a single period, which is the reason for preferring a bounded family; the fact that draws can be negative is not itself a defect, since a failed search is discarded by the over the innovation draw, the imitation draw, and the incumbent technology. The Beta shape parameters are not calibrated to patent-value data. The patent-value literature (Harhoff et al. 1999 and successors) is qualitative motivation for choosing a skewed, bounded step distribution; it does not pin down and , which are set to keep innovations mostly incremental and are then subject to sensitivity analysis.
Embodied technical change is load-bearing
The assumption that productivity gains are embodied in capital goods (machines) rather than floating as disembodied TFP is structurally important. It means that a firm must invest to access better technology. If the firm cannot invest - because demand is weak, credit is tight, or internal funds are exhausted - it is stuck with old machines. This creates the Keynesian-Schumpeterian linkage: demand conditions affect investment, investment affects technology adoption, and technology adoption affects productivity growth. Remove embodiment and the link breaks.
Transition checks
Agent update
The aggregate is computed after individual states are updated. If the rule is nonlinear or stochastic, the aggregate can change abruptly even when the plotted line connects adjacent periods.
Each point is an aggregate after one model update. The connecting line helps the eye, but the law is discrete.
Do not read the line as a continuous differential equation unless one is stated.
Cutoff and contagion
The indicator changes value discontinuously at . Aggregates inherit that discontinuity when enough agents sit near the cutoff.
A cutoff creates a kink or jump because the update rule changes at the threshold.
Do not force a smooth explanation onto a chart section driven by threshold crossing.
Variable timing
Agent state
Agent states update at dates. Abrupt aggregate movement can come from many small state changes applied at once.
Aggregate readout
The aggregate can jump when the distribution changes shape or many agents cross a boundary.
Threshold indicator
The indicator turns on at the cutoff. That creates a discontinuity in the micro rule and possibly the aggregate.