| dc.description.abstract | The curse of dimensionality is frequently encountered in applied time series econometrics when incorporating information in large datasets. Here, three applications are presented that faced challenges in dimensionality and were resolved differently: by either the a priori placement of restrictions on the parameter space or using data-driven techniques. One such data-driven method is the extraction of latent factors to reduce the information contained in many variables into a smaller number of series. A second data-driven method is statistical inference on conditional correlations to infer causality.
A parsimonious model of adoption is applied to study a large dataset of adoptions of the ductless heat pump (DHP), an energy-efficient technology. This research aims to increase the understanding of DHP adoption in the Pacific Northwest of the US by quantifying the effect of utility-provided rebates and Northwest Energy Efficiency Alliance (NEEA) expenditures on the number of installations and providing forecasts of DHP installations through 2018 given various rebate and NEEA expenditure levels. NEEA desires to increase installations of DHPs by providing funding for marketing and training; however, forecasted installations through 2018 do not meet their goals. Adoptions of DHPs are elastic with respect to the net cost of installation, and a reduction of federal tax rebates in 2011 decreases the probability of adoption.
The second objective is to investigate the dynamic effects of shocks in oil supply, aggregate demand, and oil demand on oil prices, the upstream, midstream, and downstream sectors of the petroleum industry, and the broader US economy. Because
neither the petroleum industry nor the economy can be described in a small number of series, the analysis is performed in a data-rich environment, applying a time-varying parameter extension to the factor-augmented vector autoregression. Results suggest the effects of shocks in oil supply, aggregate demand, and oil demand have evolved, and oil supply shocks play a larger role in the dynamics of the petroleum industry during recessions.
The final objective is to investigate the appropriateness of the PC Algorithm as a subset vector autoregression (VAR) methodology in determining both the contemporaneous and lag structure of the data-generating process. Subset VARs might improve forecasts and/or allow for the inclusion of more time series through a reduction in parameterization. Monte Carlo experiments show the PC Algorithm is effective at discovering the lag and contemporaneous structure of a VAR. Additional observations increase the algorithm’s efficacy. When researchers do not have a priori knowledge of the true number of lags in the data-generating process, overfitting provides a smaller penalty than underfitting. | en |
