Analysis of U.S. Housing Markets Using Advanced Econometric Models and Machine Learning Algorithms
Date of Award
Doctor of Philosophy
Matthew Higgins, Ph.D.
Eskander Alvi, Ph.D.
Kevin Corder, Ph.D.
Econometrics, forecasting, great recession, housing, machine learning, monetary policy
The Great Recession is the most recent reminder in U.S. history of the important relationship that exists between monetary policy and housing market activity. Since World War II, eight U.S. recessions have followed downturns in housing. A decline in the housing market results in the balance sheet deterioration of homeowners, financial institutions, private organizations, central banks, governing bodies, and foreign countries. The ability to understand the relationship between monetary policy and housing market activity, and to better forecast housing prices, allows for better planning by all economic agents.
This study answers the following research questions. (1) How have monetary policy shocks altered housing market responses during the conventional and unconventional monetary policy periods? Monthly data from the Federal Reserve is split into two sample periods corresponding to the two different monetary policy regimes observed during the Great Recession. Bayesian Vector Autoregressive models are estimated to study the sign restricted impulse response functions of different housing market indicators to shocks in the federal funds rate and the Federal Reserve’s total asset holdings. Results indicate that residential investment declines in the wake of a contractionary monetary policy shock during the conventional policy period. However, no statistically significant response is detected during the unconventional monetary policy period. (2) What would have happened to housing prices if the Federal Reserve failed to carry out any of the unconventional monetary policy measures that were observed during the Great Recession? I estimate separate Structural Vector Autoregressive models based on the excess demand and constrained supply theories to produce different counterfactual simulations through a historical decomposition methodology. Historical decompositions indicate that both theories are needed to reproduce the observed housing price series. However, the theory of constrained supply best reconstructs the housing price decline observed during the onset of the Great Recession. Results from the counterfactual simulations indicate that average housing prices would have depreciated by an additional $28,200 if not for the unconventional asset purchases made by the Federal Reserve. (3) Does more data result in better housing price forecasts? Three data sets of increasing size are created from multiple sources to estimate 31 different econometric models and machine learning algorithms in order to forecast real housing price growth rates at the U.S. national and Census regional levels. Although results differ by geographical region, simple univariate models tend to produce forecasts comparable to the models built on large information sets. Additionally, the small information set models, which are designed to mimic the small econometric models commonly studied in the literature, result in the weakest improvement in predictive accuracy.
In totality, the research presented in this study highlights the need for agents to continuously question the long-established economic assumptions in the aftermath of the Great Recession.
Restricted to Campus until
Kirby, Cody, "Analysis of U.S. Housing Markets Using Advanced Econometric Models and Machine Learning Algorithms" (2023). Dissertations. 3940.