Modeling Fiscal Stimulus Effects with Causal Impact Methods
Introduction
Government fiscal stimulus—tax cuts, spending increases, or transfers—aims to stimulate economic growth and employment. However, measuring the true causal impact is notoriously difficult; correlation between stimulus announcements and subsequent growth does not imply causation, as other factors evolve simultaneously. Causal inference methods, particularly Bayesian causal impact analysis, allow investors to isolate the true effect of fiscal stimulus on asset prices and economic outcomes, improving forecasting accuracy and informing positioning around fiscal announcements.
Challenges in Estimating Fiscal Stimulus Effects
Endogeneity and Reverse Causality
Fiscal stimulus is often announced during recessions, when growth is already weakening. Naive regression of growth on stimulus shows positive correlation, but this conflates stimulus effect with mean reversion in growth. Markets anticipate both the stimulus and underlying economic dynamics, making simple regression unreliable for estimating marginal impact.
Confounding Variables
When stimulus is announced, multiple other factors change simultaneously: Fed expectations adjust, consumer sentiment shifts, business investment responds. Disentangling stimulus effects from these confounding influences requires sophisticated econometric techniques.
Bayesian Causal Impact Framework
Counterfactual Construction
The core idea: estimate what would have happened absent stimulus (the "counterfactual" outcome), then compare to actual outcomes. The difference is the causal effect. Bayesian Structural Time-Series (BSTS) models achieve this by fitting a statistical model to pre-stimulus data, then extrapolating to estimate counterfactual post-stimulus outcomes.
Synthetic Controls
A more general approach: identify a set of "control" assets or regions not affected by stimulus (or less affected), then use their post-stimulus performance to model what would have happened without stimulus. For example, when estimating US fiscal stimulus effects, international equity markets (less directly affected) serve as controls. The difference between US and international returns represents the stimulus causal effect.
Implementation for Asset Price Analysis
Case Study: COVID Fiscal Stimulus (2020)
In March 2020, the US announced unprecedented fiscal relief (CARES Act: $2 trillion). Simultaneously, equity markets were in free fall and the Fed launched emergency programs. To isolate fiscal effects, construct a counterfactual using:
- Developed market equity returns (Europe, Japan) as control
- Pre-announcement (January–February 2020) relationship between US and global equity returns
- The estimated relationship extrapolated to April–June 2020 generates counterfactual US returns
- Actual minus counterfactual returns reveal fiscal stimulus causal effect
Multi-Asset Causal Analysis
Extend beyond equities to bonds and currencies:
- Fiscal stimulus increases government debt, raising real yields; compare to counterfactual based on global bond yields
- Stimulus increases inflation expectations, weakening currency; compare to counterfactual based on commodity currencies
Data and Econometric Methodology
Bayesian Structural Time-Series Models
BSTS models decompose asset returns into components:
- Trend (underlying long-run direction)
- Seasonality (recurring patterns)
- Exogenous regressors (Fed policy, equity valuations)
- Error terms
Causal Impact Package and Software
Google's CausalImpact (R and Python ports available) implements this methodology with minimal code. Researchers specify:
- Pre-period (data used for model fitting)
- Post-period (period whose causal effect is estimated)
- Response variable (e.g., asset return) and control variables
Portfolio Applications
Fiscal Policy Trading Strategies
Estimate causal impact of proposed stimulus on equities, bonds, and currencies. If the estimated effect suggests strong fiscal stimulus should boost equities but markets are pricing weak stimulus impact, tactical overweight equities captures alpha. Conversely, overpriced fiscal effects suggest taking profits or shifting to defensive positioning.
Risk Management
Understanding causal impacts improves scenario analysis. In stress tests, quantify how different stimulus levels (no stimulus, modest package, large package) causally affect portfolio returns. This informs hedging decisions and risk capital allocation.
Central Bank Policy Interaction
Causal impact analysis reveals how fiscal stimulus effects depend on concurrent monetary policy. Large stimulus with accommodative Fed amplifies effects; large stimulus with tightening Fed offsets effects. Understanding these interactions improves nowcasting and policy response prediction.
Challenges and Limitations
Limited Treatment Variation
Fiscal stimulus is not randomly assigned; there are only a handful of major stimulus episodes. This limits ability to estimate heterogeneous treatment effects or assess how stimulus effects vary across economic regimes. Careful study design and sensitivity analysis help address this constraint.
Model Specification
Causal impact estimates are sensitive to choice of control variables and model specification. Robustness checks—varying control sets, model structures, pre-period windows—ensure results are stable and not driven by arbitrary choices.
Conclusion
Causal inference methods, particularly Bayesian causal impact analysis, enable investors to isolate true effects of fiscal stimulus on asset prices from confounding factors. By carefully constructing counterfactuals using control variables and Bayesian time-series models, institutional investors can make more informed decisions about fiscal stimulus positioning, avoiding over- and under-reactions to stimulus announcements. As fiscal policy becomes increasingly deployed alongside monetary policy, causal impact analysis is essential for sophisticated portfolio management.