Date of Award




Document Type


Degree Name

Doctor of Philosophy (PhD)


Department of Economics

Content Description

1 online resource (xi, 150 pages) : illustrations (some color)

Dissertation/Thesis Chair

Kajal Lahiri

Committee Members

Ulrich Hounyo, Zhongwen Liang


Forecasting, Recession, Tax Revenues, Tax revenue estimating, Recessions, Econometric models

Subject Categories



In recent years models with mixed frequency have been extensively used to forecast low-frequency variables such as GDP and inflation, but we are one of the first to use this framework in state government revenue forecasting. New York State had a record of passing late budgets before. In order to facilitate budget negotiations, which often center on forecasts, we develop a Mixed-Data Sampling (MIDAS) model for revenue forecasting using jagged edge data sets in the first chapter. We forecast yearly tax revenues using monthly data on tax receipts and also two dynamic factors extracted from a set of selected monthly and quarterly indicators specific to the New York State and the U.S. economy separately. These three models are combined with optimal weights to generate monthly multi-period forecasts. The weights of the two dynamic factors are high at horizons more than 11 months, after which the monthly tax revenue variable picks up in its contribution as uncertainty is resolved. By combining we gain forecast efficiency at all horizons. Our sample covers fiscal years 1986-2020; and data till FY 2007 is used in estimation to generate and evaluate out-of-sample forecasts over FY 2008-2020. To coincide with the budget process, our forecasts start 18 months, and are continuously updated monthly till the end of the fiscal year. Our model allows for identification of reasons for forecast revisions as new information arrives on a monthly basis in a transparent manner. We document significant gains in forecast accuracy. The relative gain in forecasting efficiency is particularly significant during the cyclical downturns, including COVID-19 in 2020. Counterfactual analysis is implemented to study the marginal contribution of data at each horizon. We estimate the variances of combined out-of-sample forecasts with blocking-based residual bootstrap methodology. With the variance estimates, we provide fan charts for each fixed target fiscal year showing the underlying forecast uncertainty.

Included in

Economics Commons