Printer Friendly

Searching for a better forecast: San Francisco's revenue forecasting model.

Placing forecasting in the context of the local economy resulted in more accurate revenue projections for San Francisco and a better understanding of the forces determining the city's available resources.

In public-sector budgeting, the availability of resources usually circumscribes discussions about expenditures. As these discussions intensify in the face of mounting fiscal duress, reliable and informative revenue forecasts become critical elements of the budgeting process. This is increasingly the case for all state and local governments, and has been particularly true for the city and county of San Francisco since 1988.

In the spring of 1988, San Francisco faced the daunting task of closing an unprecedented general fund deficit - for FY1989 the general fund deficit was projected at $192 million, an amount equal to 17 percent of the prior year's general fund budget. Understandably, the size of the deficit prompted the mayor's office to carefully review all revenue and expenditure estimates, in order to understand the underlying causes of the problem and fashion the mayor's budget proposal accordingly.

The Need for a Better Model

In the course of this review it became clear that the city's method for revenue forecasting was inadequate. For example, there was little systematic reliance on national, regional or local economic variables to inform the forecasts. Consequently, economic trends or events affecting the underlying tax base of the city, such as changes in employment, had little impact on, or relevance for, the forecasts. In addition, isolating trends which were particularly relevant when discussing tax increases, or new levies, to alleviate the deficit was impossible. Instead, these trends and events simply became components of the anecdotal library which was, in reality, the city's principal forecasting tool. Most importantly, the revenue forecasts were off from actual receipts by wider and wider margins.

Given the rising discomfort over these revenue forecasts, and the prospect of another deficit in FY1990, the mayor's office endeavored to create a forecasting tool which would * improve the accuracy of the city's

forecasts, * enable the mayor's office to analyze and

identify economic trends and events

shaping the underlying tax base and * be relatively easy to use.

By placing forecasting in the context of the local economy, the mayor's office hoped to gain a better understanding of the forces determining the city's available resources and improve the accuracy of the forecasts. Accurate and informed forecasts also offered the prospect of greatly enhancing decision making during the budget process.

Developing a model that was relatively inexpensive to create and easy to use was also an important consideration. The model would have to be developed so that it could run on software compatible with an Apple Macintosh 512K personal computer, the only data processing resource available to the mayor's office. The data which would be used in generating these forecasts had to be readily accessible from inexpensive government publications, since staff had no funds available for purchasing large data bases. The variables selected for use in the model would have to be easily available from newspapers and magazines, so as not to incur additional costs for specialized data. With these goals and constraints in mind, the financial analyst in the mayor's office set out to build a revenue forecasting model.

Components of the New Model

San Francisco's revenue model is a set of 23 econometric equations, which are divided into three discrete, yet interrelated, components. The first component consists of 10 equations which forecast San Francisco's underlying tax base, defined here as output. The second component consists of five equations to forecast key local or regional economic variables, such as San Francisco's total assessed value. The final component consists of eight equations to forecast the revenues generated by seven local taxes and one state subvention. The local taxes are property, business, sales and use, utility user, property transfer, parking and transient occupancy. The one subvention is the motor-vehicle-in-lieu payment, which is a sort of property tax on motor vehicles. In total, these eight forecasted revenue sources will comprise approximately 65 percent of the city's total general fund resources of $1.35 billion in FY1992.

Identifying the Underlying Tax Base. Most taxes are directly or indirectly levied on some aspect of economic activity, whether it be the appreciation in property values, the sale of goods and services or some other events. The taxed economic activity is simply a manifestation of the overall national, regional or local economy that forms the underlying tax base upon which governments levy their taxes. Identifying and modelling San Francisco's economy, consequently, became central to any attempt to forecast tax receipts.

The first component of San Francisco's revenue model seeks to identify the city's underlying tax base by creating a simulation of the San Francisco economy based on a model developed for the New York metropolitan area by Matthew P. Drennan.(1) Drennan's model employs a major theoretical concept in urban and regional economics - economic base theory - to identify the underlying tax base of a region.

Economic base theory argues that an economy's industries fall into two categories: "basic" (or export) industries export goods and services outside of the local or regional economy, thereby generating income from other parts of the nation; "nonbasic" (or local) industries produce goods and services which are consumed within the local economy. Because of their importance to the local or regional economy, basic industries are thought to indirectly create additional jobs and income in the local economy. This multiplier effect refers to the impact that an additional dollar expended or job created in an export or basic industry has on the local or nonbasic industries.(2)

Drennan outlines a method whereby basic and nonbasic industries could be identified and their output measured. Using national labor statistics and national income measured by the U.S. Census Bureau's Standard Industrial Classifications (SIC), a time series of annual "value added per employee" (VAPE) for each nonagricultural industry is created. This annual VAPE by industry is then multiplied by the number of local employees in each industry for the corresponding time period to estimate annual output by industry for a particular region. Export industries are then identified by comparing the ratio of an industry's local output to total output for all local industries with the ratio of that industry's national output to total output in all national industries. (These ratios are referred to as "location quotients" in urban economics.) Those local industries which have a higher ratio than the national ratio for that industry are said to be exporting this additional output to the rest of the nation. Exhibit 1 illustrates how Drennan's methodology was used to create the San Francisco output variable.


After identifying basic and nonbasic industries, the derived output measures are used to specify econometric equations (one for each of the industries), which predict future annual output generated by each industry in that region. Each industry's equation contains a variety of variables which are statistically significant predictors of output for a given industry. Basic industry equations rely on national variables, principally measures of general activity in the national economy, such as changes in gross national product (GNP). Nonbasic industry equations rely on measures of activity in the local economy, principally the output generated by the basic industries and the multiplier effect this output has on the "nonbasic" industries.

While there were many elements of Drennan's model which could not be replicated for San Francisco, because of limited staff resources - changes in the relevant data and differences between the two economies - a model of the city's economy finally emerged in the summer of 1988. This model of the underlying tax base uses Drennan's methodology to develop output measures for the nine non-agricultural single-digit SIC industries from 1972 through 1989. These output data were then used to specify statistically significant equations to predict each industry's output. Of the nine industries, transportation and utilities, wholesale trade, finance, insurance, real estate (FIRE) and services were identified as export industries over that 18-year period.(3) Mining, construction, manufacturing, retail trade and government were identified as local industries. A tenth equation, forecasting total local industry income using total export industry income, also was specified in order to derive an intermediate total output variable needed to forecast output in each of the local industries.

Key Regional and Local Variables. The mayor's office, having identified the underlying San Francisco tax base using Drennan's methodology, proceeded to specify revenue equations which utilized this San Francisco output variable, the relevant tax rates and any exemptions which were part of the administration of the tax. This approach worked for five of the revenues. In two of these instances, the property tax and the motor-vehicle-in-lieu subvention, it became clear that a second set of regional and local variables, which constituted the actual tax base for each of these revenues, would be necessary.

This second component of the model consists of five equations which are used to forecast key local or regional economic variables used in the revenue forecasts for property taxes and the motor-vehicle-in-lieu subvention, or provide additional information on the local economy. These variables are: personal income for the state of California, population for the state of California, total assessed value in San Francisco, personal income for San Francisco and total population for San Francisco. Demand in the national economy, measured by changes in GNP, demand in the local economy, as measured by output in the San Francisco economy, the interest rates of three-month U.S. Treasury bills and the consumer price indices for the San Francisco Bay area and the nation are used to produce forecasts of these variables.

The Revenue Forecasts. The final component of the model consists of the eight equations used to generate the actual forecasts for the revenues highlighted earlier. Each equation contains a set of variables which attempt to capture: * the appropriate tax base output

modified to reflect any exemptions

granted to industries; in the case of the

property, motor-vehicle-in-lieu and hotel

equations, a more relevant tax base is

utilized: * the relevant tax rate; and * a variable to capture exemptions of

particular components of the tax base

(e.g., the utility users' tax exempts

residential users) or one-time events

(such as the 1989 earthquake) which

may have affected receipts in a given

year. In most cases, the revenue source was deflated by the GNP deflator in order to remove the effects of inflation and better gauge the consequences of changes in the tax rates, the aforementioned exemptions and one-time policy changes.

An interesting challenge arose in the course of specifying the revenue equations: matching the time frame in which the economic output was generated with the taxes collected by the city. San Francisco, like most governments, utilizes a fiscal year that does not begin January 1; however, the output data, was constructed on a calendar year basis. Since there is a real delay between the actual creation of wealth by the private sector and the payment of taxes on this wealth, a one-year lag was utilized in all of the equations, except the property tax equation. That is to say, the economic output of the calendar year preceding the fiscal year of interest was used to forecast revenues; e.g., calendar year 1990 data were used to forecast FY1991 revenues. The property tax equation utilizes a two-year lag.

Has the Model Been Useful?

The mayor's office had identified three goals it hoped to achieve by building a revenue model: 1) improve the accuracy of the city's revenue forecasts; 2) provide a policy tool to analyze the city's tax and revenue policy; and 3) accomplish the preceding without significant cost and with relative ease. The model which was finally developed accomplished each of these goals with some success.

Were the Forecasts Accurate? Expecting any model to provide forecasts which are accurate to the last dollar is unreasonable, but model builders hope that forecasts will be within a reasonable margin of error, usually within 5 percent of actual receipts. While in San Francisco these diminished expectations were kept in mind, the accuracy of the forecasts was still surprising, as Exhibit 2 illustrates. In FY1989 - the first year of the model's use - forecasts of the eight revenues underpredicted actual receipts by a total of $845,000 or 0.1 percent. FY1990 actual receipts totalled $803.6 million, while the forecasts anticipated a total of $797.6 million, a difference of $5.9 million, or 0.7 percent. In FY1991 the variance was 1.9 percent; the forecast overestimating receipts in this year.


Two comments should be made about the model's performance in the first two years. The first is that property transfer taxes have been (and continue to be) the most difficult revenues to predict. In all years, the forecasts for property transfer taxes missed the actual receipts by unacceptable margins; underpredicting by 15.1 percent in FY1989 and overpredicting by 6.5 percent in FY1990 and by 45 percent in FY1991. Most troubling about the property transfer tax variance is that it is inconsistent and therefore difficult to correct. Surveys of other municipal revenue officers confirmed that this was also their most difficult revenue source to forecast. Overall, the forecasts have been relatively accurate when the property transfer tax forecast is removed from the aggregate forecast. This performance suggests two things: that the model can capture the impact of downturns in the economy and that attempts to forecast property transfer taxes may not be possible using this tool. This may be because property transfers are subject to factors, such as the availability of foreign capital, changes in bank regulations and consumer psychology, which are difficult, if not impossible, to capture using the economic variables included in the model.

The second comment concerns the FY1990 forecasts. In October 1989 San Francisco suffered a severe earthquake which made revenue forecasting very difficult. Aside from exacerbating the aforementioned difficulties associated with forecasting property transfer taxes, the earthquake also affected forecasts for property taxes and utility users' taxes. The underprediction of property taxes appears to have been the result of one-time tax relief provided by the State of California in FY1990, and a brief suspension of assessment and tax payment rules. The earthquake also may have caused businesses in San Francisco to consume less energy and communication services than normal, since it took two to three weeks for the utility companies to restore services to preearthquake levels. Attempts to model these effects of the earthquake proved inconclusive.

Were the Forecasts Informative? The revenue model also provided the mayor's office with an informative tool for tax and policy analysis. The model enabled the staff to develop realistic estimates of the revenue impact of tax rate changes, the effect of these rate changes on output and employment and the impact of various economic development proposals on the city's revenues. For example, the model became a critical tool for analyzing the potential impact of the mayor's proposal to build a new facility for the San Francisco Giants baseball team, since it enabled a thorough analysis of the revenue implications of keeping the team in San Francisco. (4) Without the model, this kind of analysis would have been impossible.

Is the Model Easy to Use? Although the model was difficult to develop, it has proven to be very easy to use. The equations have been entered onto a Microsoft Excel spreadsheet which allows staff to continuously update the forecasts, as well as test various scenarios regarding economic growth. For example, analysts can provide the director of finance with monthly forecasts during budget deliberations by simply entering new values onto the spreadsheet for the variables of interest.

The cost of developing the model, aside from staff time needed to develop it, was minimal. The statistical package which was used to estimate the equations, Stat Works, was purchased for less than $200. All of the relevant data were garnered from government documents available at minimal cost or from the public library. The spreadsheet application for the personal computer was already available in the mayor's office.

Gaining acceptance of this revenue model as a legitimate tool for forecasting and tax and policy analysis has been difficult. The controller's office, which was primarily responsible for producing revenue forecasts before the creation of the mayor's office revenue model, initially resisted using any of the forecasts generated by the model. Acceptance of the model's forecasts has come grudgingly and only recently, spurred by the model's performance and the need to have as reliable an estimate of available resources as possible. In the FY1992 budget deliberations the model was used as the basis of all budget decisions regarding revenues.

Resource constraints, technological limitations, gaps in knowledge and bureaucratic comfort with existing standard operating procedures are formidable hurdles that often make innovation in the public sector difficult. In San Francisco these hurdles were encountered as the mayor's office sought to improve the information utilized during budget formulation. Funding was scarce; computing capabilities were nonexistent; knowledge about the economy was cursory; and existing standard operating procedures proved difficult to alter.

Despite the obstacles, development of the model has given the city an important policy tool. Decisions on the allocation of scarce resources are more informed because the mayor and staff are better knowledgeable about the extent of these resources.

San Francisco's experience with revenue modeling underscores the importance of innovation in the public sector, particularly at the local level. In this era of scare local resources - and the high opportunity costs associated with utilizing these resources unwisely - innovation may be one of the few means by which local governments can do more, qualitatively, if not quantitatively, with less.


(1) Modelling Metropolitan Economies for Forecasting and Policy Analysis, (New York and London: New York University Press, 1985). (2) Although economic base theory is accepted as a very useful method for analyzing urban and regional economies, a number if urban economists argue that input/output modelling, which seeks to capture the value added to the economy by a particular industry, presents a more accurate "snapshot" of the value of economic output. For a variety of reasons, principally the costs associated with creating an input/output matrix or purchasing a series of these matrices from the Bureau of Economic Analysis, the mayor's office decided to pursue a model which utilized economic base theory as the theoretical framework. James Heilbrun provides an excellent discussion of the relative merits and problems of each of these theories in Urban Economics and Public Policy, Third Edition, (New York: St. Martin's Press, 1987) pp. 139-171. (3) Since 1980 wholesale trade has not been an export industry. Aside from representing a significant change in San Francisco's economy, this made forecasting this industry's output problematic. Since wholesale trade has been an export industry for most of the time period, the model treats it as a basic industry for each year. (4) See Wilkins, Carol and Agostini, Steven J., "Building a New Home for the San Francisco Giants: A Cost-Benefit Analysis of the Proposed China Basin Ballpark," October 2, 1989, a public report available from the mayor's office in San Francisco or from the author.

Stephen J. Agostini, is executive policy and program analyst for the office of the Executive Director of The Port Authority of New York and New Jersey. He was formerly co-budget director/economic and financial analyst in the mayor's office, city and county of San Francisco. Readers wishing additional information on the design and operation of the forecasting model can contact the author at 212/435-2401.
COPYRIGHT 1991 Government Finance Officers Association
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1991 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Agostini, Stephen J.
Publication:Government Finance Review
Date:Dec 1, 1991
Previous Article:Automating assessment information for improvement districts: the City of Tempe's award-winning system.
Next Article:Economic indicators: vital signs monitor economy's health.

Related Articles
Mapping default zones.
Scouting Top Apartment Markets.
State revenue forecasting: An institutional framework.

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters