Grounding Large Language Models in DSGE Simulators for Policy Generation and Forecasting

Aditya Dubey; Namah Gupta; Vinti Agarwal

cs.AIOct 1, 2026

Grounding Large Language Models in DSGE Simulators for Policy Generation and Forecasting

Authors: Aditya Dubey, Namah Gupta, Vinti Agarwal

Organizations: Birla Institute of Technology and Science, Pilani, India

Abstract

Large language models can produce economic policy responses that sound reasonable, but this does not show that their actions are consistent with economic dynamics. We test this by placing an instruction-tuned language model inside six Snowdrop-backed dynamic stochastic general equilibrium (DSGE) simulators. At each turn, the model observes the economy and a change in economic discourse, selects a bounded policy action, and receives the next simulated state and an economic reward. We implement a common Python interface for repeated rollouts, persistent shocks, state cloning, and rolling-horizon simulation. This setting creates a long-horizon credit-assignment problem. Policy effects may appear several quarters after an action is taken. PPO has a learned value function that can propagate delayed reward to earlier tokens through generalized advantage estimation. GRPO has no learned value function and instead assigns a group-relative advantage from complete rollout returns. It therefore cannot distinguish which earlier turn caused the outcome; if every rollout receives the same return, the normalized advantage is zero. We use PPO as the primary method and GRPO as a matched critic-free baseline. The experiments also test directional semantic signals, reward horizon, trajectory warm starts, cross-simulator transfer, and historically anchored pandemic and monetary-policy shocks. The objective is to judge policy actions by their simulated economic consequences rather than by plausible language alone.

Figures & tables

Figure 1: Semantic grounding architecture. Economic discourse is encoded into a temporal semantic state, mapped to named shock hypotheses, and evaluated by a DSGE simulator rather than by linguistic plausibility alone.

Figure 2: Closed-loop policy grounding. A semantic shift conditions the policy agent, a validated action modifies the DSGE shock path, and multivariate simulator outputs determine the reward and next observation.

Key	Structural model	Principal targets	Policy lever(s)
QPM	Quarterly projection	inflation, output gap	monetary, fiscal
SW	Smets–Wouters	inflation, output	monetary, fiscal
GSW	Galí–SW pandemic	inflation, output, unemployment	monetary, fiscal
IRE	Ireland (2004)	inflation, output gap	monetary
MVF	U.S. multivariate filter	inflation, output, unemployment	demand stabilizer
RBC	Real business cycle	output, consumption	output stabilizer

Table 1: Structural models exposed through the common environment.

Figure 3: Paired financial narratives with small but decision-relevant semantic differences. Adapted from the Financial-STS examples of Liu et al. (2024) ; the examples motivate supervision beyond static topic similarity.

Source	Contents	Role	Primary metrics
DSGE registry	controlled shocks, states, discourse	training and ablations	return, loss, validity
COVID-19 release	scenario shocks and G-Cubed outcomes	structural stress test	RMSE, correlation, loss improvement
ECB/EA-MPD	meeting and monthly policy surprises	historical event validation	sign, timing, stabilization loss

Table 2: Data sources and their non-overlapping roles in the evaluation.

Model/test	Period	Simulated variables
QPM, $SHK\_RS=1.0$		$RS$	$RR$	$DLA\_CPI$
	Initial	16.010000	8.461200	7.250700
	Next quarter	13.272388	11.004444	5.525716
SW, multiple shocks		$y$	$pinf$	$r$
	Initial	0.000000	0.000000	0.000000
	Next quarter	$-0.035428$	$-0.007073$	0.020122

Table 3: Representative outputs from the ten-case simulator verification suite. Values are reproduced from the recorded QPM and SW verification output.

Horizon	Method	Return	Econ. loss	Valid action	Credit diagnostic
Short	Base LLM	TBD	TBD	TBD	–
Short	GRPO	TBD	TBD	TBD	TBD
Short	PPO	TBD	TBD	TBD	TBD
Long	Base LLM	TBD	TBD	TBD	–
Long	GRPO	TBD	TBD	TBD	TBD
Long	PPO	TBD	TBD	TBD	TBD

Table 4: Confirmatory PPO–GRPO comparison. Values will be inserted from saved rerun artifacts; lower economic loss and higher return are better.

Question	Primary contrast	Hypothesized direction
Semantic shift	directional vs. static text	higher shock F1; lower loss
Reward horizon	rolling vs. one-step	fewer policy-ranking reversals
Reward geometry	tolerance band vs. exact target	greater target-zone occupancy
Exploration	warm-start vs. direct RL	more useful actions; lower final loss
Transfer	multi-model vs. single-model	lower held-out-model loss

Table 5: Directional summary of the confirmatory ablations. Numeric estimates and confidence intervals remain to be populated from the rerun.

Scenario / action	Magnitude	Weighting	Base loss	Action loss	$\Delta$	Direction
Recession / cut	0.05	Standard	216	219	$-3.0$	worse
Recession / cut	0.50	Standard	216	248	$-32.0$	worse
Stagflation / hike	0.05	Hawk	334	333	$+1.6$	better
Stagflation / hike	0.50	Standard	369	378	$-9.6$	worse
Cost-push / hike	0.25	Standard	2664	2996	$-331.8$	worse

Table 6: Calibration grid search from the initial DSGE experiments.

Explore similar work

Sep 4, 2026cs.CL

Can Large Language Models Anticipate Behavioral Responses to Social Policies? A Case of Pension Enrollment Prediction among China's Flexible Workers

Assessing the impacts of social policy changes is a widely acknowledged challenge for policymakers. Econometric methods can be unreliable when extrapolating to hypothetical scenarios, while field pilot programs are highly costly. In this paper, we propose using large language models (LLMs) as policy-assessment tools adapted from general-purpose models. We present FlexPension-LLM, the first domain-specialized large language model for a hierarchical pension-enrollment prediction task among flexible workers in China, and introduce DKI-RDistill, which injects policy-grounded cues into the prompt, including Probit-derived marginal effects and hukou-province pension rules. The method then uses LoRA/SFT to distill rationale-augmented supervision into an open-weight MoE student, with teacher errors corrected by regenerating those cases under ground-truth labels. On a CHFS 2019 blind split, FlexPension-LLM achieves 0.9316 Composite F1, surpassing its Claude Sonnet 4.5 teacher and 15 of 17 baselines, and is statistically indistinguishable from Claude Opus 4.6. Across four external surveys, it averages 0.7549 Composite F1 and shows the narrowest performance range among the strongest systems. Component analysis shows that gains come mainly from policy-grounded cue injection and error-filtered supervision, while rationales provide decision traces that can be checked against policy rules.

Yumiao Li, Peixin Liu, Donglin Di +2

School of Economics and Management, Tsinghua University, Beijing, China · Actuarial Science and Risk Management, University of Illinois Urbana-Champaign, Urbana, IL, USA · School of Computing, Harbin Institute of Technology, Harbin, China

Large Language Model Policy Optimization Agent Behavior Modeling Contextual Cues

Ask PDF

Apr 16, 2026cs.AI

Meituan Merchant Business Diagnosis via Policy-Guided Dual-Process User Simulation

Simulating group-level user behavior enables scalable counterfactual evaluation of merchant strategies without costly online experiments. However, building a trustworthy simulator faces two structural challenges. First, information incompleteness causes reasoning-based simulators to over-rationalize when unobserved factors such as offline context and implicit habits are missing. Second, mechanism duality requires capturing both interpretable preferences and implicit statistical regularities, which no single paradigm achieves alone. We propose Policy-Guided Hybrid Simulation (PGHS), a dual-process framework that mines transferable decision policies from behavioral trajectories and uses them as a shared alignment layer. This layer anchors an LLM-based reasoning branch that prevents over-rationalization and an ML-based fitting branch that absorbs implicit regularities. Group-level predictions from both branches are fused for complementary correction. We deploy PGHS on Meituan with 101 merchants and over 26,000 trajectories. PGHS achieves a group simulation error of 8.80%, improving over the best reasoning-based and fitting-based baselines by 45.8% and 40.9% respectively.

Ziyang Chen, Renbing Chen, Daowei Li +4

LongCat Interaction Team, Meituan Inc. Beijing, China · Independent Researcher Changsha, China

User Simulation Simulation-Based Reinforcement Learning Behavioral Foundation Models

Ask PDF

Jun 18, 2026econ.GN

AI Economist Agent: An Agentic Framework for Evidence-Based Economic and Financial Analysis with RAG, Knowledge Graphs, and Large Language Models

We propose an AI economist agent for economic and financial scenario analysis. Scenario design often requires analysts to assess emerging risks with limited historical precedent, combine information from many sources, and translate qualitative mechanisms into internally consistent quantitative paths. Large language models (LLMs) can search and synthesize this information, but fluent narratives alone do not establish the model-based calculations needed for economic conclusions. Our framework uses LLM agents to plan the analysis, retrieve relevant evidence, and organize economic mechanisms, while registered quantitative models generate numerical outcomes and predefined tests determine whether intermediate results can be used in the final report. We apply the framework to European macro-financial stress scenarios and bank capital analysis. The empirical analysis evaluates retrieval of economic mechanisms, scenario construction, model execution, and report generation under a historical information cutoff. The results show how the AI economist agent can combine flexible evidence retrieval and scenario construction while keeping the resulting analysis linked to identifiable sources and explicit model calculations.

Masahiro Kato

The University of Tokyo

Economies Data Science Agents Narrative Generation Research Automation Report Agentic Framework+1

Ask PDF