美联储-互换反击：对一年通胀预期的评估（英）-2023.9-43页_1mb

The Swaps Strike Back: Evaluating Inflation Expectations

This paper examines the relative forecasting performance of one-year inflation expectations derived from inflation swaps and survey-based measures (Blue Chip Economic Indicators), focusing on their accuracy and identifying periods where one method dominates the other. Excluding episodes of severe liquidity issues (e.g., during the Global Financial Crisis and early pandemic), inflation swaps perform better overall in predicting realized inflation and show superior performance under specific conditions.

Key findings:

🎯 Core Results

• Inflation Swaps vs. Surveys:
  • Full sample analysis shows surveys are slightly better using MSE but swaps are closer to realized inflation in 56% of months and perform better in "clean" samples that exclude liquidity crisis periods.
  • Surveys tend to reflect mode forecasts (rigid to biases), while swaps reflect probability-weighted means and have higher variance (~1.76% mean vs surveys' 2.14%).

Conditions Where Swaps Outperform:

• When inflation above its historical median (2% since 2004), swaps significantly outperform.
• When the gap (swaps - surveys) > -30 basis points (median gap), swaps also lag fewer steps.

🧮 Forecast Combination Strategies

Static Weighting:

• An equal-weight combination (0.5 swaps + 0.5 surveys) improves forecasts but dynamic roll-windows overweight swaps (~0.7 since ~2020).

Dynamic Smooth-Transition Model:

• Weight inflation swaps higher when its level >1%, making it costly during low/deflation expectations due to liquidity fears~→ eliminates redundancy during "normal" times. Named DCS measure, it reduces error significantly across conditioning variables.

📉 Caveats

• Short sample period limits retrospective-out performance comparison.
• No factor adjustment controls reduce swap performance.
• Realized inflation outperforms proxies in volatility ranges.

💎 Conclusion

Inflation swaps are cleaner indicators than surveys for out-of-sample inflation projections outside liquidity shocks. Optimal forecasts combine both methods, yet DCS model provides strong performance as a regime-dependent hybrid, especially during economic duress.