Estimating Labor Force and Fiscal Modules for Coastal Louisiana Economies: Extension of the COMPAS Modeling Framework Presented by: Arun Adhikari Dr. J Matthew Fannin Projected Funded Through Cooperative Agreement with Minerals Management Service, US Dept of Interior

Outline  Introduction and Objective of the Study  Background and Overview of COMPAS Modeling  Data and Methodology  Empirical Specifications  Models Discussion  Results and Discussion  Concluding Remarks

Introduction  Accuracy in any policy analysis is of a great value to public decision makers  This study also aims to develop a model to forecast different expenditure demands in the fiscal module of the Louisiana Community Impact Model (LCIM) using alternative procedures capable of increasing the performance over traditional COMPAS estimators.  The specific objective includes modeling the fiscal module (four major categories of expenditure; public service, public works, general government and health and welfare) of LCIM for 60 parishes of Louisiana to compare the performance between spatial and non spatial estimators that takes into account heterogeneity.

Background and Overview of COMPAS Modeling  COMPAS models are regional economic models that combine input-output and econometric approaches to build a conjoined model of economic structure.  COMPAS models typically treat employment demand as an exogenous driver of changes in the labor market which ultimately impact the fiscal sector.  The fiscal module in this research is an extension to the module used by Fannin et al., (2008). The goal of this analysis is to adhere to the basic theme of regional science: spatial location matters.

Background and Overview of COMPAS Modeling

Data  Estimation is based on the COMPAS model for Louisiana that includes 60 parishes, where the variables for the fiscal module were selected on the basis of Fannin et al (2008) and were modified depending upon the requirements of our model and applied geographically to all Louisiana parishes.  Data sources: a)Audited Financial Statements b)Bureau of Economic Analysis c)U.S. Census Bureau d) Department of Education  Within the fiscal module, different expenditure equation data on public safety, public works, general government and health and welfare sectors were estimated.

Methodology  Four expenditure equations were estimated. Every equation is a function of several specific variables from all Louisiana parishes.  These equations were estimated by a cross-section Ordinary Least Square (OLS) model as a base control with quantile regression, and spatial autoregressive model regressions also estimated.  We applied OLS regression and quantile regression using STATA, and spatial regression using MATLAB.  Base year of estimation is 2007.

Empirical Specifications  Four different expenditure equations that were estimated for comparison are: 1)lnpcgg = f (lnpcasdv, lnpcretsl, lnpcin, lnpurb, lnarblndnsty) 1)lnpchw = f (lnpcasdv, lnpcretsl, lnpcin, lnpafam) 1)lnpcps = f (lnpcasdv, lnpcretsl, lnpcin, lnpafam, lnarblndnsty) 1)lnpcgg = f (lnpcasdv, lnpcretsl, lnpcin, lnpurb, lnarblndnsty, lnpclcrdml)

Performances Evaluation and Comparison  Forecast performances were evaluated based on the procedures outlined in Johnson, Otto and Deller (2006), and Kovalyova and Johnson (2006).  The performance of estimators is compared on the basis of quantitative evaluation methods. These methods include analysis of: a)mean simulation error (ME), b)mean percent error (MPE), c)mean absolute error (MAE), d)mean absolute percent error (MAPE), e)mean square error (MSE), f)root mean square error (RMSE), g)root mean square percent error (RMSPE), h)and Theil’s coefficient U1 and U2

Models Discussion 1)Ordinary Least Squares (OLS) 2) Quantile Regressions  Divided into 3 quantiles (0.33, 0.66 and 0.99) 3) Spatial Autoregressive Regression (SAR)  Derivation of spatial weight matrix

Results Table 1: Variable description and summary statistics, Louisiana, 2007 Variable NameMeanStandard DeviationMinMax General Government Expenditure9,176,81929,103,832593,955210,722,026 Health and Welfare Expenditure2,125,6582,924,0595,66413,602,439 Public Safety Expenditure7,008,96525,111,070232,882189,130,903 Public Works Expenditure9,549,50711,714,106847,07065,739,927 Assessed Value418,151,563553,860,43936,056,8643,466,560,930 Retail Sales901,353,1451,355,501,80929,883,9467,612,001,075 Arable Land Density ,909 Local Road Miles1, ,635 Population60,19174,1235,828429,914 Per Capita Income2,87395,21820,06043,206 Percent African American Percent Urban

Results Table 4: Performance Estimation of 66 th quantile for Public Safety, Louisiana, 2007 b1b2b3b4b5b AreanameyYhatyhat-yAbs (Yhat-y)(yhat-y)/yAbs (Yhat-y)/y(yhat-y) 2 {(yhat-y)/y} 2 y2y2 yhat 2 Catahoula Red River Livingston Iberia Vernon Jackson Bossier Jefferson Davis Bienville Beauregard Ascension Tensas Caldwell Webster DeSoto Vermillion West Feliciana Natchitoches Tangipahoa East Carroll SUM Avrg MSE1, RMSE36.76 U10.07 U20.52

Results Table 3: Average performance estimation measures for different categories of expenditure, Louisiana, 2007 Expenditure CategorySpatial AutoregressiveLinear (OLS)Quantile Regression General Government yhat-y (yhat-y)/y {(yhat-y)/y} Theil’s Coeff (U1) Health and Welfare yhat-y (yhat-y)/y {(yhat-y)/y} Theil’s Coeff (U1)

Expenditure Category Spatial Autoregressive Linear (OLS)Quantile Regression Coeff.t-statCoeff.t-statCoeff.t-statCoeff.t-statCoeff.t-stat General Government Constant-20.24*** * ** **-2.19 percapasdval0.54** *** * * percapretsls percapinc1.95*** * ** **2.16 percenturban * arblndensity * *1.78 rho-0.49***-4.16 Health and Welfare Constant-28.57*** ** ** *** percapasdval percapretsls percapinc2.73*** ** * percentafam ** *** rho Results Table 2: Parameter estimates for OLS and Quantile regressions, Louisiana, 2007

Major Results  An increase in assessed value leads to increase in the expenditure of the all categories, as seen from all three models.  An increase in per capita income leads to increase in expenditure in the public safety for all the categories of expenditure, as seen from all three models. The magnitude keeps increasing for higher quantiles.  Lower and median quantiles are found to be performing better as compared to spatial and OLS regressions.  Some parishes like Ascension, Bienville, Iberia and West Feliciana are not performing as good on average. On the contrary, parishes like Catahoula, Jefferson parish, Red river and Vermillion are performing better than the average error measures.

Concluding Remarks  This research sets out a strategy for comparing the forecasting performances between spatial (SAR) and non-spatial models (OLS and Quantile Regressions) in a fiscal sector of LCIM.  Evaluation of alternative methodologies are expected to give regional economic modelers better information from which to choose when seeking to construct models projecting different modules.  These results will be helpful to those community modelers desiring to estimate cross-section fiscal modules for forecasting in states that have much greater heterogeneity among local government units.  Other spatial models like the spatial error model and panel data models could also be evaluated while comparing the performances between spatial and non-spatial estimators, which would be a concept for future research in this paper.

Thank You Questions/Comments