Climate-Macro ABM Model

Integrated assessment models (IAMs) like Nordhaus's DICE (1992) and RICE treat the economy as a single representative agent optimizing consumption against a climate damage function. That compression erases three things that matter for climate-macro policy. First, the distribution of carbon intensity across firms: a carbon tax that raises the average cost by 5% might bankrupt the dirtiest 10% of firms while leaving clean firms unaffected, and the aggregate GDP effect depends on whether those bankruptcies cascade through supply chains and bank balance sheets. Second, the endogenous dynamics of green innovation: firms do not switch technologies instantaneously in response to a price signal - they invest, experiment, fail, and diffuse innovations through networks, and the speed of that process depends on the current distribution of technological capabilities as well as the carbon price level. Third, the financial system's role in amplifying or dampening transition risk: banks holding loans to carbon-intensive firms face correlated losses when policy tightens, and their response (tightening credit to all firms, including clean firms) can turn an orderly transition into a recession. Climate-macro ABMs exist to capture these three channels simultaneously.

The field crystallized around 2015-2020. Lamperti, Dosi, and colleagues at the Sant'Anna School built the DSK (Dystopian SchumpKy) model (2018, 2020), extending the Keynes+Schumpeter (K+S) macro ABM with a climate module: firms emit carbon as a byproduct of production, emissions accumulate in the atmosphere, atmospheric CO2 drives temperature change via a simplified climate model, and temperature anomalies cause stochastic damage to firm capital and household income. The key finding was that climate-induced financial instability - banks failing because their borrowers suffer climate damages - amplifies GDP losses by 15-30% beyond the direct damage channel. Lamperti et al. (2019) added green transition dynamics: firms can invest in green R&D to reduce emissions intensity, and the innovation process follows a stochastic search over a technology space. Ponta et al. (2018) built a separate climate-macro ABM focused on energy sector transition with heterogeneous power generators. Mercure et al. (2018) developed the FTT (Future Technology Transformations) model for global energy transition with evolutionary technology dynamics. Monasterolo and Raberto (2018) constructed EIRIN, a stock-flow consistent climate-macro ABM focused on the financial system's exposure to climate transition risk.

Central banks and financial regulators have picked up climate-macro ABMs for stress testing. The Network for Greening the Financial System (NGFS), which coordinates climate risk analysis across 130+ central banks, has published scenario frameworks that draw on ABM-derived insights about nonlinear transition risk. The ECB's climate stress test (2022) used scenario paths informed by models that capture firm heterogeneity in carbon exposure. The Bank of England's climate biennial exploratory scenario (CBES, 2021) included transition risk channels where carbon-intensive firm failures propagate through the banking system - a mechanism that only ABMs model endogenously. Academic use spans from pure climate-economy theory (Nordhaus critics like Keen, Farmer, Hepburn) to applied green industrial policy evaluation (subsidy design, carbon border adjustments, green bond markets).

The economy is assembled from five agent populations. Firms are the primary agents: each carries a production technology characterized by labor productivity and carbon intensity (emissions per unit output), a capital stock subject to climate damage, a balance sheet (debt, equity, retained earnings), and behavioral rules for production, pricing, investment, and R&D. Firms are split into a capital goods sector (producing machines of varying carbon intensity) and a consumption goods sector (using those machines to produce output). The firm population is typically 200 to 500 in research implementations. Households form the second population: each carries income (wages or unemployment benefits), wealth, and a consumption rule. Some variants give households heterogeneous climate preferences that affect demand for green versus brown goods. Banks form the third population: each holds a loan portfolio with exposure to firms of varying carbon intensity, a capital buffer, and lending rules. The government is the fourth agent: it sets the carbon tax, redistributes carbon tax revenue, funds green subsidies, and runs fiscal policy. The fifth agent is the climate module - not a decision-making agent but a physical system that accumulates CO2, computes radiative forcing, and maps temperature anomalies to stochastic damage shocks on firm capital and productivity.

Interaction happens through four channels. The goods market connects firms and households: firms post prices (cost-plus markup adjusted for carbon tax), households allocate demand based on price and, in some variants, green preference. The labor market connects firms and households: firms hire workers, pay wages, and fire when demand falls. The credit market connects firms and banks: firms apply for loans to finance investment and R&D, banks screen based on firm financials and (optionally) carbon exposure. The innovation network connects capital goods firms: green R&D investment yields stochastic improvements in carbon intensity with probability depending on current technological distance from the frontier, and successful innovations diffuse through capital goods sales to consumption goods firms. The climate module sits underneath all four channels, imposing damage shocks whose severity depends on cumulative emissions.

State variables update each period in a fixed sequence: (1) the climate module updates atmospheric CO2 and temperature given last period's emissions; (2) climate damage shocks hit firm capital and productivity; (3) firms make production, pricing, investment, and R&D decisions; (4) the goods market clears; (5) the labor market adjusts; (6) firms apply for credit, banks screen and lend; (7) firm profits and losses are realized, defaults occur, bank capital absorbs losses; (8) the government collects carbon tax revenue, redistributes, adjusts policy; (9) aggregate statistics are recorded and emissions from this period's production are sent to the climate module. This sequential structure makes the propagation path from a carbon price increase through firm costs, investment reallocation, bank exposure shifts, and aggregate output traceable step by step.

Lamperti et al. (2020) used the DSK model to quantify the public costs of climate-induced financial instability. Running Monte Carlo simulations under IPCC RCP scenarios, they found that climate damages transmitted through the banking system amplify GDP losses by 15-30% beyond the direct damage channel. A firm hit by a climate shock defaults on its bank loan; the bank absorbs the loss, tightens lending to damaged and undamaged firms, credit-constrained firms cut investment, output falls further, more firms become fragile, and the next climate shock hits a weaker economy. This amplification channel is invisible in DICE-type models because there is no banking sector to transmit and amplify shocks. The finding directly influenced NGFS scenario design, which now includes financial amplification as a distinct risk channel separate from physical and transition risk.

Lamperti et al. (2019) used the green transition variant to evaluate carbon tax design. They compared a flat carbon tax (same rate for all firms) against a revenue-recycling carbon tax (revenue returned as green R&D subsidies) against a ramp-up carbon tax (low initial rate increasing over 20 years). The revenue-recycling variant dominated on both GDP and emissions metrics: it maintained demand (no net fiscal drag) while accelerating green innovation, achieving 40% emissions reduction with only 2% GDP cost over 30 years, compared to 35% emissions reduction and 5% GDP cost for the flat tax. The ramp-up tax achieved the deepest long-run emissions cuts (50%) but at higher short-run transition cost because firms delayed green investment waiting for the price signal to strengthen. These results depend on the firm heterogeneity and innovation dynamics that the ABM captures - a representative-agent IAM cannot distinguish between these policy designs because it has no firm distribution to redistribute toward.

The model breaks down in several settings. Pure physical climate risk assessment (what is the expected damage to a specific asset from sea-level rise or extreme weather) is better served by catastrophe models and climate science, not a macro ABM. Very long-run (century-scale) projections push the model beyond its calibration window and accumulate compounding errors from the simplified climate module. Economies where the carbon-intensive sector is small relative to GDP (e.g., service-dominated economies with low manufacturing shares) show weak transition risk amplification because the banking exposure channel has low volume. For these settings, a DSGE with a climate externality (Golosov et al. 2014) or a detailed energy system model (TIMES, REMIND) may be more appropriate.