Data-Rich DSGE Model Forecasts of the Great Recession and its Recovery 

Review of Economic Dynamics--Volume 32, April 2019, pp.18-41

I investigate the extent to which modern dynamic stochastic general equilibrium (DSGE) models can produce macroeconomic and labor market dynamics in response to a financial crisis that are consistent with the experience of the Great Recession. Using the methods of Boivin and Giannoni (2006) and Kryshko (2011), I estimate two DSGE models in a data-rich environment. The two models estimated in this paper include close variations of the Smets & Wouters (2003, 2007) New Keynesian model and the FRBNY (Del Negro et al. 2013) model that augments the Smets & Wouters model with a financial accelerator. I find the model with a financial accelerator that is estimated in a data-rich environment is able to significantly out-forecast modern DSGE models not estimated in a data-rich environment and the Survey of Professional Forecasters (SPF) in regard to core macroeconomic growth variables and many labor and financial metrics including the unemployment rate, total number of employees by sector and business loans. 

Working Paper Version

Online Appendix

Previous Title: Financial Crises, Recoveries and Labor Market Dynamics: Evidence from a Data-Rich DSGE Model 


The Effects of Professional Forecast Dissemination on Macroeconomic Volatility

Journal of Economic Behavior and Organization--Volume 170, February 2020, pp.131-156

This paper explores the role that professional forecast announcements can have on macroeconomic volatility. Bounded rational agents are used inside a medium scale dynamic stochastic general equilibrium (DSGE) model with financial frictions. Modeled agents must form expectations about endogenous variables by selecting between three simple linear forecasting specifications some of which contain the inclusion of a“professionally announced forecast of the economic variable. Historically calibrated simulations of the model show that the usage of the announced professional forecast by the agents in their adaptive learning forecast specifications can reduce the volatility in consumption, inflation and wages by as much as 26%, 23% and 22% respectively. However, if the professional forecast is not disseminated well to the agents or biased in its dissemination, agents will learn to ignore the announcement and macroeconomic volatility will increase. Further, the inclusion of very noisy professional forecast signals can result in “coordinated volatility cascades” where agents could reduce macroeconomic volatility by ignoring the professional forecast but choose not to because of its previous forecast performance. 


Re-evaluating Okun’s Law: Why all recessions and recoveries are "different"

Economics Letters--Volume 196, November 2020, 109497 

This paper explores the relationship between GDP growth, the unemployment rate and employment growth. Using a structural DSGE model and a data-rich estimation approach I am able to estimate the coefficients and correlations between GDP growth and unemployment rate changes, GDP growth and overall employment growth as well as GDP growth and employment growth by sector. I find historically equivalent estimates when I compare the simulated model with actual realized data. I am then able to look at the effect different types of economic shocks have on these estimates and I find that investment and finance shocks have larger effects on employment growth and the unemployment rate when compared to productivity and other supply-side shocks when the effect on GDP is controlled for. For example, the model suggests that a 1% decline in real GDP would result in an increase of the unemployment rate of 0.5% if it was caused by a financial or investment shock, however, it would only increase the unemployment rate by 0.15% if the 1% decline in real GDP was caused by a productivity shock.

Working Paper

Examining Business Cycles and Optimal Monetary Policy in a Regional DSGE Model

In this paper, I construct a dynamic stochastic general equilibrium (DSGE) model consisting of geographic regions and use state level data to estimate the effects that monetary policy and financial shocks have on the four census regions of the United States. The DSGE model I use is constructed around a centralized monetary authority and financial market with regional output, labor and investment markets and is a close variant of the FRBNY model (Del Negro et al. 2013). I use a combination of state level and national level data to estimate the regional and national parameters of the DSGE model. I find significant heterogeneity amongst the regional structural parameters of the model, creating different dynamics for the four regions in regard to national monetary and financial shocks. Simulating the estimated model, I find that monetary policy that considers the regional variation in output and inflation can significantly lower a central bank’s loss function while also being Pareto improving to all four regions. The paper’s results suggest that regional macroeconomic conditions should be considered in monetary policy decisions.


Working Paper

Evaluating the Forecasting Power of an open-economy DSGE model when estimated in a Data-Rich Environment

This paper examines the inferences and forecasting benefits that can be made when one incorporates a large quantity of economic time series into international structural macroeconomic models. I estimate a close variation of Adolfson et al. (2007, 2008) small open-economy dynamic stochastic general equilibrium (DSGE) model in a data- rich environment and evaluate its predictive performance of the Canadian macroeconomy. The data set I use in the paper includes Canadian, American, Asian and European macro-financial data. I compare the forecasting performance of the DSGE model estimated in a data-rich environment (DSGE-DFM) to the forecasts generated by the DSGE model under estimated in its traditional setting and forecasts generated by other reduce formed forecasting models. I find that an open-economy DSGE model estimated in a data-rich environment significantly out performs its regularly estimated DSGE counterpart and the DSGE-DFM forecasts that incorporate real-time data are similar or better to the Bank of Canada’s Staff Economic Projections for GDP, consumption, investment, and trade statistics. In addition, the DSGE-DFM model of this paper is useful in forecasting both the real and nominal exchange rate in the short and medium-term.

Work in Progress

Large-scale Macroeconometric Models vs International DSGE-DFM models

joint with Chris Gibbs (University of Sydney), Dan Rees (Bank of International Settlements) and Luke Hartigan (Reserve Bank of Australia)

The primary objective of this project is to compare International DSGE-DFM models to large-scale macroeconometric (LSM) models. Specifically, we want to know whether DSGE-DFM models forecast key domestic and international macroeconomic aggregates as well as LSM models and whether the two models provide consistent predictions for the effect of monetary policy.

Working Paper

Considering the State of the World when conducting Real-Time Optimal Pool (RTOP) Model Weighting


I compare the out-of-sample forecasting performance of the two Dynamic Stochastic General Equilibrium (DSGE) models outlined in Gelfer (2019) when they are estimated both out of and in a data-rich environment. I find that the DSGE model with a financial accelerator out forecasts the model without financial frictions in only two periods over the first decade of the 21st century. I also find that the DGSE models estimated in a data rich environment (DSGE-DFM) significantly out forecast their regularly estimated counterpart in regard to output, consumption, and investment growth. In the second half of the paper I evaluate the out-of-sample forecasting performance of these four models against that of other forecasting models including VAR models, dynamic factor models, and structural FAVAR models. Additionally,  I explore the “value” of model averaging DSGE-DFM models in terms of forecasting performance and I find that accounting for what type of micro-finance data is currently most volatile in real-time optimal pool model weighting procedures,  produces the best out-of-sample forecasts for the entire forecast window (1999-2013) for all core macroeconomic production variables and inflation.