Key Takeaways

• Bayesian forecasting is currently leading the field of macroeconomic forecasting.
• Simple models often outperform complex ones in macro forecasting.
• Advancements in big data and machine learning have significantly evolved forecasting models.
• Overfitting is a constant risk in econometric models, posing challenges for forecasters.
• Policymakers need a narrative behind forecasts to understand economic mechanisms.
• Simpler models are more effective for policymakers due to clearer storytelling.
• Bayesian econometrics is popular in the US for its ability to impose priors on data.
• Bayesian methods are advantageous in data-scarce situations.
• AI enhances forecasting by processing vast data but introduces interpretation challenges.
• High-frequency data can add noise during stable periods but is valuable during significant changes.
• AI forecasts using historical data may not perform well in real-time situations.
• The FOMC’s economic projections influence market reactions significantly.
• Market reactions differ between FOMC meetings with and without economic projections.
• The September FOMC meetings provide a noisy proxy for the Fed’s true forecast.
• Surprises in monetary policy are larger during September FOMC meetings.

Guest intro

Andrew Martinez is an assistant professor of economics at American University. He previously served as an economist in the Office of Macroeconomic Analysis at the US Department of the Treasury. His research specializes in time series econometrics and forecasting.

The evolution of macro forecasting

• Bayesian forecasting is at the forefront of macroeconomic forecasting today.

• Bayesian methods have evolved significantly over the last several decades.

  — Andrew Martinez

• Simple models often outperform complex ones in macro forecasting.

• It’s really hard to often beat very simple models.

  — Andrew Martinez

• Big data and machine learning have transformed forecasting models.

• Machine learning and model selection algorithms have really come across.

  — Andrew Martinez

• Overfitting is a major challenge in econometric models.

• There’s constantly this risk of overfitting.

  — Andrew Martinez

• Policymakers require a narrative to understand forecasts.

• It’s important to have that story.

  — Andrew Martinez

The role of Bayesian econometrics

• Bayesian econometrics allows for imposing priors, useful in data-scarce situations.

• The value of Bayesian is that you’re able to impose these priors on the data.

  — Andrew Martinez

• Bayesian methods are popular in the US for econometrics.

• Bayesian econometrics is a very popular school of thought.

  — Andrew Martinez

• Bayesian thinking involves updating beliefs with new data.

• Updating prior beliefs with new data is central to Bayesian thinking.

  — Andrew Martinez

• Relying too much on priors can lead to biased results.

• If you’re relying too much on those priors, it can lead to biased results.

  — Andrew Martinez

AI’s impact on economic forecasting

• AI can process vast amounts of data, enhancing forecasting.

• AI is very useful at accessing all of our publicly available information.

  — Andrew Martinez

• AI introduces challenges in data interpretation.

• AI can introduce challenges in data interpretation.

  — Andrew Martinez

• High-frequency data provides insights but can add noise during stable periods.

• High-frequency data can add noise when not going through a major shift.

  — Andrew Martinez

• AI forecasts may not perform well in real-time situations.

• AI forecasts look amazing historically but struggle in real-time.

  — Andrew Martinez

The significance of FOMC meetings

• FOMC economic projections significantly influence market reactions.

• FOMC projections play a crucial role in market reactions.

  — Andrew Martinez

• Market reactions differ between FOMC meetings with and without projections.

• Markets react differently to non-SEP releases versus SEP releases.

  — Andrew Martinez

• September meetings provide a noisy proxy for the Fed’s true forecast.

• The September meetings are a noisy measure of the Fed’s true forecast.

  — Andrew Martinez

• Surprises in monetary policy are larger during September meetings.

• Surprises are significantly larger during SEP release meetings.

  — Andrew Martinez

The expectations gap and GDP measures

• The expectations gap is constructed by comparing forecasts to actual GDP outcomes.

• Comparing forecasts to actual outcomes creates the expectations gap.

  — Andrew Martinez

• The Hamilton filter has been extended for better output gap measures.

• The Hamilton filter has been extended for improved accuracy.

  — Andrew Martinez

• Empirical output gap measures can be applied to nominal and real GDP.

• Output gap measures can be applied to both nominal and real GDP.

  — Andrew Martinez

• There is a lack of interest in nominal GDP measures despite their usefulness.

• Nominal GDP measures have been overlooked but are very useful.

  — Andrew Martinez

The challenges of real-time forecasting

• AI models struggle with real-time data due to biases in interpreting historical information.

• AI sometimes biases itself with historical data.

  — Andrew Martinez

• AI is trained on a mix of structured and unstructured data, complicating real-time insights.

• AI’s mix of data complicates its real-time accuracy.

  — Andrew Martinez

• Policymakers and forecasters face difficulties with limited real-time information.

• Limited information sets challenge policymakers and forecasters.

  — Andrew Martinez

• Real-time data constraints affect forecasting accuracy.

• Real-time data constraints impact forecasting accuracy.

  — Andrew Martinez

The influence of monetary policy shocks

• Monetary policy shocks can have unexpected responses due to additional information.

• Central bank communications release additional information affecting responses.

  — Andrew Martinez

• Information effects influence monetary policy responses beyond shocks.

• Additional information effects influence policy responses.

  — Andrew Martinez

• The nominal GDP gap measures the difference between expected and actual GDP.

• The nominal GDP gap assesses aggregate demand pressures.

  — Andrew Martinez

• Information effects persist in monetary policy after controlling for factors.

• Information effects persist even after controlling for various factors.

  — Andrew Martinez

The importance of adaptive forecasting models

• Forecasting models should adapt to changing economic conditions.

• Models should adapt rather than rely on naive linear trends.

  — Andrew Martinez

• Averaging forecasts provides a robust measure of economic conditions.

• Averaging forecasts is simple and theory-free.

  — Andrew Martinez

• Forecast-based GDP measures are more reliable than traditional output gaps.

• Forecast-based measures are less volatile and more reliable.

  — Andrew Martinez

• The expectations gap from Fed forecasts is less subject to revision.

• The expectations gap is much less subject to revision.

  — Andrew Martinez