Forecasting Growth During the Great Recession: Is Financial Volatility the Missing Ingredient?

Laurent Ferrara, Clément Marsilli, Juan‐Pablo OrtegaSSRN 2013 · DOI

TL;DR

This paper investigates whether financial volatility, specifically from stock and commodity markets, can improve GDP growth forecasts during economic downturns like the Great Recession. The authors develop an extended Mixed Data Sampling (MIDAS) model that effectively combines daily financial data with monthly economic indicators, demonstrating that financial volatility significantly enhances forecasting accuracy, especially for longer horizons.

What Problem Does This Paper Address?

The Great Recession of 2008-2009 highlighted the crucial role of financial markets in driving macroeconomic fluctuations. While much research has explored the link between financial variables and economic cycles, there's a relative scarcity of studies focusing on the predictive power of financial volatility for macroeconomic outcomes. A key challenge in this area is integrating high-frequency financial data (daily) with lower-frequency macroeconomic data (quarterly GDP) without losing valuable information through temporal aggregation. This paper aims to bridge this gap by proposing a method to effectively combine these disparate data frequencies to improve GDP growth forecasting, particularly during periods of economic stress.

Methods

The core of the paper's methodology is an extended Mixed Data Sampling (MIDAS) model. MIDAS is designed to handle forecasting when explanatory variables are sampled at different frequencies than the target variable. In this case, the target is quarterly GDP growth, while explanatory variables include daily financial volatilities and monthly industrial production.

The authors extend the standard MIDAS framework to incorporate:

  • Daily Financial Volatility: They estimate daily volatilities for commodity prices (CRB index) and stock market indices (S&P500, CAC40, FTSE100) using an AR(1)-GARCH(1,1) model. This is a standard approach for modeling financial volatility.
  • Monthly Industrial Production: This serves as a traditional macroeconomic leading indicator.
  • Direct Multi-Step Forecasting: Instead of forecasting one step ahead and iterating, they estimate a separate model for each forecasting horizon (from 0 to 11/3 quarters ahead). This approach is chosen to mitigate the impact of specification errors in one-step-ahead models.

The paper compares three models:

  1. Model M<sub>d</sub>: Uses only daily financial volatilities.
  2. Model M<sub>m</sub>: Uses only monthly industrial production.
  3. Model M<sub>dm</sub>: Combines both daily financial volatilities and monthly industrial production.

These models are benchmarked against a simple AR(1) model that uses only past GDP values.

Datasets

The study utilizes publicly available data for three major economies: the United States (US), France, and the United Kingdom (UK).

  • Quarterly Output: Real GDP growth rates.
  • US: 1976q1-2010q4
    • France: 1988q1-2010q4
    • UK: 1988q1-2010q4
  • Daily Volatilities: Derived from:
  • CRB spot price index (Commodity Research Bureau)
    • S&P500 index (US)
    • CAC40 index (France)
    • FTSE100 index (UK)
  • Monthly Series: Industrial Production Manufacturing Index (IPI) for each country.

The out-of-sample forecasting experience covers the period 2007q1-2010q4, which crucially includes the Great Recession.

Key Insights and Ideas

  • Financial Volatility Matters for Forecasting: The study provides strong evidence that daily financial volatility, particularly from commodity and stock markets, is a valuable predictor of future GDP growth, even beyond traditional macroeconomic indicators.
  • MIDAS is Key for Mixed Frequencies: The MIDAS approach is essential for effectively integrating daily financial data with quarterly GDP data without losing information, a common pitfall with simpler aggregation methods.
  • Combining Sources is Superior: The model that combines both financial volatilities and industrial production (Model M<sub>dm</sub>) consistently outperforms models that use either source alone or just past GDP. This underscores the benefit of a multi-faceted approach to forecasting.
  • Horizon-Specific Importance: The relative importance of financial variables versus industrial production shifts with the forecasting horizon. Financial volatilities tend to be more impactful for longer horizons, while industrial production gains importance closer to the release date of the GDP data.
  • The Great Recession as a Stress Test: The study's focus on the Great Recession period demonstrates that the predictive power of financial volatility is particularly pronounced during times of economic turmoil and uncertainty.

Key Results

The paper's main findings are presented through Root Mean Square Forecasting Errors (RMSFE) for various forecasting horizons.

  • Model M<sub>dm</sub> Dominance: Across all three countries (US, France, UK) and for most forecasting horizons, the combined model (M<sub>dm</sub>) yields the lowest RMSFE, indicating superior forecasting accuracy.
  • Significant Gains from Financial Volatility: For the US and UK, Model M<sub>d</sub> (financial volatility only) often performs comparably to or better than Model M<sub>m</sub> (industrial production only) for longer horizons (e.g., 4/3 to 7/3 quarters ahead), suggesting that industrial production alone is not sufficient for these lead times.
  • Specific Country Nuances: While M<sub>dm</sub> is generally best, the UK results show Model M<sub>d</sub> leading the combined model for some horizons, suggesting a particularly strong role for financial variables in forecasting the UK economy. For France and the US, industrial production's impact becomes more pronounced closer to the release date.
  • No Predictive Gain from Industrial Production Alone for Long Horizons: For horizons of 4 to 7 months before the target date, the paper notes that including industrial production does not add predictive value beyond what financial information already provides.

Limitations and Future Work

  • Data Availability: The study acknowledges that the availability of high-frequency financial data can be an issue, and that handling such data requires careful consideration (e.g., overnight effects, price misrecordings).
  • Model Specificity: The study focuses on specific financial variables (commodity and stock prices) and a particular GARCH model for volatility estimation. Future work could explore other financial assets or alternative volatility estimation techniques.
  • Forecasting Horizon Insights: While the paper identifies interesting patterns in the relative importance of variables across horizons, further research could delve deeper into the economic mechanisms driving these shifts.
  • Generalizability: While the findings are robust across three major economies, extending the analysis to other countries or different economic regimes could provide broader insights.

Summary updated

BibTeX
@misc{ferrara2013forecasting,
  author = {Laurent Ferrara and Clément Marsilli and Juan‐Pablo Ortega},
  title = {{Forecasting Growth During the Great Recession: Is Financial Volatility the Missing Ingredient?}},
  year = {2013},
  doi = {10.2139/ssrn.2343132},
  howpublished = {SSRN},
}
Figure