Keith R. Phillips

Seasonal adjustment of hybrid time series: An application to U.S. regional jobs data

Hybrid time series data often require special care in estimating seasonal factors. Series such as the state and metro area Current Employment Statistics produced by the U.S. Bureau of Labor Statistics (BLS) are composed of two different source series that often have two different seasonal patterns. In this paper we address the process to test for differing seasonal patterns within the hybrid series. We also discuss how to apply differing seasonal factors to the separate parts of the hybrid series. Currently, for state employment data, the BLS simply juxtaposes the two different sets of seasonal factors at the transition point between the benchmark part of the data and the survey part. We argue that the seasonal factors should be extrapolated at the transition point or that an adjustment should be made to the level of the unadjusted data to correct for a bias in the survey part of the data caused by differing seasonal factors at the transition month.

An evaluation of real-time forecasting performance across 10 western U.S. States

The recent financial crisis and economic downturn has emphasized the importance of accurate subnational forecasting models. To judge which models work best, researchers have emphasized the importance of looking at the true real-time performance of models and not simply an analysis of out-of-sample results. In this study we utilize real-time forecasts from the Western Blue Chip Economic Forecast to analyze and evaluate a host of different forecasters and models across time and 10 U.S. states to see if some models and forecasters consistently outperform others. We use the forecast accuracy criteria established by the Blue Chip publication. To evaluate accuracy we develop a scoring procedure based on the number of years that the forecaster/model was closest to actual relative to what we would expect just by random chance. We also utilize standard measures such as the Root Mean Square Error and Theil’s inequality coefficient and test the statistical significance of the best forecasts. We then take a closer look at one model that has proven to be very accurate.

Nowcasting regional job growth using payroll processing company data

Data from large payroll processing companies are a potential source of regional job growth estimates that are timelier than official government statistics. Monthly private sector employment data from ADP, LLC is available for 29 U.S. states and Washington DC about 10 days prior to official employment data from the U.S. Bureau of Labor Statistics. Using a simple model with real-time data, lags of job growth and the coincident month ADP estimates, we find that for all 30 regions the information in the ADP is statistically significant in nowcasting the most recent month of job growth. Our results suggest that the ADP data are an important source of information for analysts conducting regional current analysis.

Border Economic Recovery Lags Rest of State

In the recent postrecession period, economic growth on the Texas border with Mexico has been below that of the rest of Texas. We look at the factors that historically have propelled the border’s rapid economic growth and likely sources of the recent slowing relative to the rest of the state. Factors that have driven growth in previous decades, such as government spending, health care, and retail trade, have been relatively weak in recent years. Going forward, the border faces headwinds, such as weak federal government spending, and tailwinds, such as a young, growing labor force and the continued integration of the US and Mexican economies. Over the next several years, the economic performance of the border will likely improve to near or slightly above that of the state.

A note on spurious seasonal patterns and other distortions in the BLS local area unemployment statistics

This paper examines the potential sources behind statistically significant seasonal patterns in the state level seasonally adjusted Local Area Unemployment Statistics (LAUS) released by the U.S. Bureau of Labor Statistics (BLS). We find that these seasonal patterns are likely spurious and may be due to the pro-rata factors used in benchmarking the states to census regions and national totals. In addition we find that the Henderson 13 filter used by the BLS to smooth the seasonally adjusted state data often makes the data inconsistent with national labor market data. We conclude that the BLS should use seasonally adjusted data when estimating the pro-rata factors used to benchmark states to regional and national totals.

A closer look at potential distortions in state real gross domestic product: The case of the Texas energy sector

In this paper we take a closer look at a potential flaw in the measurement of Texas Real Gross Domestic Product (RGDP) – value added in the oil and gas industry. BEA estimates of Texas RGDP in oil and gas extraction have a negative correlation with factors of production and units of output. In this paper we use several different approximations of RGDP in oil and gas extraction to see which seems to be a good substitute for the BEA estimates. We find that a measure based on changes in Texas physical production of oil and gas results in an estimate of total state RGDP that is more highly correlated with Texas job growth and closer to the correlation of these measures nationally. This adjusted measure of Texas RGDP should be a better measure of Texas economic performance.