Procedural Generation of Macro-Economic Storylines for Scenario Analysis
Introduction
Scenario analysis requires plausible economic narratives: "What if inflation stays high but growth slows?" These storylines are typically written by economists using judgment. Generative models can procedurally create consistent economic narratives, generating novel scenarios that align with macroeconomic theory while remaining plausible and distinct.
Generating Consistent Macro Scenarios**
Scenario Components**
An economic scenario specifies: GDP growth, inflation, unemployment, interest rates, asset price changes, and qualitative narrative (explanation). These components must be internally consistent. A scenario with "high inflation + low rates" is economically incoherent (central banks would raise rates). Generative models must learn these constraints.
Constrained Generation**
Train a language model on historical economic narratives and corresponding macro variables. Learn the relationship: "High inflation, high unemployment, low growth → stagflation narrative." Use constraints: given a set of macro variables, generate a plausible narrative. Validate generated narratives for consistency.
Scenario Discovery and Diversity**
Exploring the Scenario Space**
Instead of hand-crafting 3-4 scenarios (bull, base, bear), use generative models to explore the space: sample different combinations of macro variables, generate scenarios, filter for coherence. Output: dozens of distinct, internally consistent scenarios spanning economic outcomes.
Novelty and Plausibility**
Generated scenarios should be novel (different from historical samples) yet plausible (aligned with macro theory and historical relationships). Use diversity penalties to avoid generating the same scenario repeatedly.
Building Scenario Ensembles**
Portfolio Testing Across Scenarios**
Generate 50 economic scenarios. For each, test portfolio performance. Analyze: in how many scenarios does the portfolio achieve target return? In how many does it breach drawdown limits? Generate scenario performance surfaces: return vs. inflation, return vs. growth, etc.
Robust Strategies**
Find allocations that perform well across the entire scenario ensemble. Optimization: maximize worst-case return across scenarios, or maximize return at specific quantile (e.g., 10th percentile scenario). Strategies robust to the full distribution of plausible futures.
Case Study: Multi-Asset Portfolio**
Asset manager generates 100 economic scenarios using a macro-aware language model. Scenarios vary: inflation (0%, 5%, 10%), growth (-2%, 1%, 4%), policy (hawkish, neutral, dovish). Each scenario generates coherent narratives and corresponding asset return assumptions.
Backtest a diversified portfolio across all scenarios. Results: portfolio achieves positive returns in 92% of scenarios, satisfies max drawdown constraint in 85%. Identify vulnerable scenarios: "high inflation + no growth + tight policy" (stagflation) challenges the portfolio. Adjust allocation to handle this tail scenario better.
Narrative Quality and Validation**
Economic Coherence**
Validate generated narratives against macro models. Does the narrative align with Phillips curve? Does the inflation path match the policy path? Use economic validators to identify incoherent scenarios.
Expert Review**
Have economists review generated scenarios. Keep plausible ones; discard incoherent ones. Use human feedback to refine the generative model.
Advanced Features**
Temporal Scenarios**
Generate not just static scenarios but paths: time-series trajectories showing how macro variables evolve over 5 years. Model transitions: recovery to boom, boom to crash, crisis to stabilization. Temporal dimension enables path-dependent portfolio decisions.
Correlation Structures**
Scenarios should include asset correlation matrices consistent with the macro environment. In stagflation, traditional diversification fails (stocks and bonds both fall). Generate scenarios with crisis correlation structures.
Limitations**
Historical Bias**
Models trained on historical data generate scenarios resembling past events. Truly novel scenarios (black swans) are rare. Augment with expert-specified extreme scenarios.
Model Risk**
The generative model encodes assumptions about macro relationships. If the model's macro theory is wrong, scenarios are misleading. Sensitivity analysis: test portfolio robustness to alternative macro theories.
Conclusion**
Procedurally generated economic scenarios enable exploration of a richer scenario space than hand-crafted ones. Portfolio managers can test robustness across dozens of plausible futures, building truly resilient strategies. The combination of macro theory, generative modeling, and expert review yields scenarios that are both rigorous and creative.