HD3 - Technical Status Report: An Overview of Expenditure Forecasting in Four Major State Programs


Executive Summary:
Nearly $15.5 billion, almost two-thirds of Virginia's general fund biennial budget of $24.7 billion, is appropriated to four State programs for the 2000-2002 biennium, as shown in the table on page II. The largest of these programs is aid to public education, which receives more than three of every ten general fund dollars. Higher education is the second largest general funded program, and Medicaid is a close third. Adult corrections is a more distant fourth, receiving slightly more than half as much as Medicaid.

Appropriations for three of these programs depend heavily upon forecasts of either the population served or expected expenditures. From 67 to nearly 100 percent of three of these large budgets (higher education being the exception) derive from forecasts of client population and unit costs. Except for higher education, these programs are "entitlement" in nature - they are required by law to serve or provide funding for services for their target populations, regardless of the size of that population. Because so much of these budgets hinge on a forecast, accuracy in forecasting is of crucial importance for budget-makers. In these cases, accurate forecasts are essential to avoid over- or under-appropriating funds. Accuracy of these forecasts in FY 1999 generally improved over prior years.

This report responds to the Appropriation Act mandate for JLARC to conduct oversight of the State's expenditure forecasting process. This mandate grew out of a legislative concern, articulated by the Joint Commission on the Commonwealth's Planning and Budgeting Process, that additional oversight was needed of key expenditure areas that increasingly dominated the State budget. The Joint Commission recommended that JLARC provide this oversight. As directed by JLARC, staff have focused on forecasts involved with the four largest general-funded programs.

Elementary and Secondary Education Enrollment Forecasting

Between 1994 and 2000, enrollment in Virginia's elementary and secondary public schools increased by 8.4 percent, to more than 1.1 million.

The State and localities share responsibility for funding public education. In the 2000-2002 biennium, the State budget provides $8 billion in general funds to direct aid for public education.

The Department of Education (DOE) is responsible for forecasting enrollment by school division each year. These forecasts are used to develop the biennial State budget, and to make adjustments to funding levels during the biennium. Enrollment is measured by average daily membership (ADM), the average number of students enrolled over the first seven months of the school year, and by fall membership, the number of students enrolled on or about September 30.

To project ADM for each school division, DOE uses simple ratios based on annual changes in ADM to fall membership ratios over the previous three years. To do this, DOE must first project fall membership levels. DOE compares the results of its projection with the fall membership forecasts independently produced by the University of Virginia's Weldon Cooper Center for Public Service. DOE generally selects the fall membership forecast closest to the Center's forecast. DOE then projects division level ADM based on historical ADM to fall membership ratios. A final step involves manual adjustments to individual division projections if localities provide information, such as unusual local trends, that would not be detected by DOE's model. This model does not look at underlying demographics or other factors, such as migration patterns.

The Center for Public Service employs a methodology that uses a grade progression ratio (based on the number of students in each grade divided by the number of students in the previous grade the year before), and conducts additional analysis to help select particular progression ratios to use in preparing a forecast.

DOE's ADM forecasts are used in the SOQ funding model to provide the number of students to be used to calculate expected costs per student for each of the 137 school divisions. These per-pupil costs are then included in the budget presented to the General Assembly each year. It should be noted that while the budget includes expected funding levels for each school division, actual payments to school divisions are based on final March 31 ADM levels.

The straightforward use of ratios in DOE forecasting has proven both accurate and easy to understand and explain. A good understanding of the process, coupled with error rates of less than one-half percent for the statewide forecast, has led to general acceptance of the forecasts by State-level budget decision-makers. Error rates at the individual division level can be much higher, however, particularly for the smaller school divisions.

Higher Education Enrollment Projections

Enrollment in Virginia's public institutions of higher education has increased nearly 13 percent over the past decade. The fall headcount for the 1999-2000 academic year was 311,536 students, with an estimated FTE level (one Full Time Equivalent - FTE - equals 15 undergraduate credit hours) of 232,348. Fall headcount is primarily used to indicate the demand on the university system, and FTEs are a factor in capital budget considerations. The higher education general fund operating budget (including the community college system) for 2000-2002 is $3.1 billion, a 15 percent increase over the prior biennium.

The State Council of Higher Education for Virginia (SCHEV) prepares a system-wide higher education budget which is publicly released prior to the Governor's biennial budget. As part of this budget process, SCHEV approves enrollment projections for the four-year institutions, in part on the basis of enrollment projections prepared by SCHEV staff. Both SCHEV and DPB project enrollment at each institution, and the institutions each project their own enrollment levels as well. The SCHEV and DPB projections are used primarily as benchmarks for assessing the reasonableness of the institutions' projections.

The higher education enrollment projection process is somewhat different from the other forecasting processes discussed in this report, because four-year institutions can be selective as to whom they provide services for, and are therefore better able to manage the size of the population they serve. For some institutions, a ''forecast" is more an indication of the institution's planned enrollment level. Another key difference is that the higher education forecasts are not directly used in calculating the State budget.

SCHEV staff develop two enrollment forecasting models for each institution. A statistical model is used to develop enrollment levels for each category of student admissions, and may employ any of a variety of statistical methods. A demographic model estimates enrollment by mapping demographic data from counties across the State to particular institutions. DPB staff use similar methodologies to develop projections. Since 1997, when SCHEV began producing six-year projections of FTE and headcount, error rates have been low. The error rates for the statewide headcount forecasts computed in the fall of 1997 were -0.4 percent for each year of the 1998-2000 biennium. The error rate in the statewide FTE projection for FY 1999 was -0.6 percent (actual FTE data is not yet available for FY 2000).

Although enrollment projections have not been strongly linked to the higher education budget process in recent years, this may change under proposals being considered by SCHEV, the Governor's Blue Ribbon Commission on Higher Education, and the Joint Subcommittee on Higher Education Funding Policies. Each of these groups is considering new funding models that could be used to bring increased uniformity to the higher education budget process, including a stronger role for enrollment projections.

Medicaid Forecasting

In FY 1999, Virginia spent $1.0 billion in general funds to provide Medicaid health care services to over 630,000 low-income Virginians. Medicaid has been the fastest growing of the four programs discussed in this report, increasing 160 percent during the 1990s, compared to an overall State general fund budget increase of 85 percent.

Two agencies, DPB and DMAS, produce independent Medicaid forecasts and then compare their results to produce an official Virginia forecast. Under language in the Appropriation Act, DPB is in effect authorized to make the final selection of a forecast upon which to base the Governor's Medicaid budget proposal.

DMAS uses two separate approaches to prepare its expenditure forecast. One approach applies exponential smoothing techniques to historical data on 70 specific Medicaid services. A second approach employs regression analysis to produce forecasts of large, acute care expenditure categories such as inpatient and outpatient hospital, physician, and pharmacy services. By combining these separate forecasts, DMAS produces a total spending forecast. For some expenditure items, DMAS staff average the results of the two approaches; for other items, the result of one or the other method is used. The expected cost of other factors, such as policy initiatives, are also included.

DPB also uses multiple methods, including regression analysis and ARIMA models, to forecast Medicaid expenditures. The final, official Medicaid forecast stems from meetings between DPB and DMAS forecasters to review data, statistical models, and technical differences. Differences are typically resolved by comparing the detailed forecasts to identify differences in assumptions or other factors. It appears that DPB staff make the final selection of forecasts to be used in preparing the biennial Medicaid budget. Two of the last three official forecasts were averages of DMAS and DPB numbers.

The FY 1999 forecasts generated by this process came within one percent of actual expenditures. The forecast adopted in the fall of 1997 for FY 1999 was 0.71 percent over the actual level of spending; the FY 1999 forecast adopted in the fall of 1998 was 0.83 percent below the actual spending level.

Inmate Population Forecasting

After several years of double-digit growth in the State's adult inmate population and a major prison construction program, Virginia has most recently seen the size of the inmate population level off and even slightly decline. The adult inmate population for which the State is responsible (under current law, felons with a sentence of one year or more) stood at 30,951 in January, 2000, down slightly from 31,181 as of June, 1999. The official forecast anticipates growth of less than three percent per year over the next ten years. This contrasts with increases of as much as 15 percent per year in the State-responsible population earlier in the 1990s.

Since the late 1980s, the Secretary of Public Safety has annually overseen a process which forecasts the number of adult inmates for whom either the State or the localities have responsibility. The forecasting process uses two committees to produce the official forecast, a technical committee that uses statistical methods (including a simulation model) to make projections, and a policy committee that reviews the projections and selects a forecast to recommend to the Secretary. Members of the policy committee include personnel who are knowledgeable about or are involved in the criminal justice process, but who are not necessarily statisticians or responsible for the incarcerated population. The policy committee also considers the effect of any newly adopted legislation on the forecast, and makes other adjustments as it deems appropriate.

The difference between the forecast and actual State-responsible inmate population in the 1996-1998 biennium ranged from 7.2 to 17.2 percent. Accuracy of the forecasts used for the 1998-2000 biennium is better; the difference between forecast and actual for FY 1999 was less than one percent. This was the only time since at least FY 1994 that the forecast used for developing the biennial budget came within five percent of the actual population.

The inclusion of additional outside parties in the inmate population forecasting process presents a useful model for the other forecasts considered in this report. Dividing the overall task between technical and policy-based issues and assigning them to appropriate personnel also brings diverse expertise to bear. This process also helps ensure that no significant trend or change is overlooked in preparing the forecast, and helps ensure a more objective forecasting result. Documentation from this forecasting process, in the form of a report issued by the Secretary of Public Safety, describes the decisions made during the process as well as the final official forecast.

Measuring Accuracy Across the Forecasts

The forecasts used to develop the 1998-2000 biennial budget were initially generated in the fall of 1997. The table on page VI compares the forecasts made during this time period for FY 1999 with the actual experience of FY 1999; these forecasts generated differences of -0.71 to +0.83 percent. Because the 1998-2000 biennial budget was adjusted again by the 1999 General Assembly, the accuracy of forecasts compiled in the fall of 1998 and used during the 1999 Session can also be assessed. Differences between forecast and actual in these mid-biennium updated forecasts for FY 1999 was closer, ranging from 0 to +0.83 percent.

While the ranges shown in the table are low, the budgetary impact of small differences can be quite high. A 0.3 percent difference in the case of elementary and secondary education enrollment, for example, led to an initial over-appropriation of $8.8 million to the Department of Education's Basic Aid program in FY 1999. In the case of the inmate population, a one percent difference could result in the development of housing for an additional 300 inmates. Despite the different impacts that can result from a small percentage difference between a forecast and the actual population or expenditure, a single standard for forecasting error may, for all forecasts, not be practical.

Although the fact that Virginia has an annual budget adjustment process tends to mitigate the need for a highly accurate biennial forecast, accuracy over a fiscal year generally is expected. This objective can be difficult and has not always been met, as illustrated by the transfer of $19.7 million from the FY 2000 Medicaid budget into FY 1999, in part due to differences between forecasted and actual expenditures. A second example was an initial FY 1999 appropriation for the ADM-based Direct Aid to Public Education accounts that was $8.8 million more than was ultimately needed.

How the Forecasts are Finalized

The process for selecting a forecast and reaching agreement that it is the most appropriate or "best" forecast is an important part of the budget process. Ideally, agreement on a particular forecast should promote agreement on the amount of funding needed to meet the forecast. The JLARC staff review found that the processes for reaching agreement in the four areas appear to be guided by a common overall strategy.

This general strategy involves comparing forecasts which are independently generated. This process can bolster confidence in the forecast that is selected, and can increase the amount of information brought to bear on the forecasting process, because it requires somewhat broader participation in the forecasting process than would otherwise be the case.

This strategy also differs from the notion that all the forecasts reviewed in this report are "consensus" forecasts. Whether multiple parties agree to a particular set of forecasts may be less important than ensuring that the forecasts are reliable and accepted for use in the budget process. Confidence in a forecast and agreement to use a particular set of numbers for budget making may be almost as important as the eventual accuracy of the selected forecast.

Documentation for three of the four forecasts is minimal, although forecasting staff provide briefings and information during the budget process. An important means of bolstering confidence may be through better documentation.

Even with a well-managed, participative, and consensus-based forecast process, there is no assurance that the resulting forecasts will be highly accurate. The State-responsible inmate population forecasts of 1996-1998 were in error by as much as 17 percent, despite participation by seven agencies and additional nonstate personnel. Ideally, broad agreement on a particular forecast should promote agreement on funding, but does not guarantee accurate budgeting.

Future Directions

This status report is the first in a series of JLARC reports dealing with forecasting in major State programs. The next report, set for late in 2000 or early in 2001, will deal more extensively with the Medicaid forecast. This will also provide JLARC staff an opportunity to respond to the new requirement, adopted by the 2000 General Assembly in SB 515 for the Department of Planning and Budget (in cooperation with the Department of Medical Assistance Services) to provide JLARC with a two-year forecast of Medicaid expenditures by November 15 of each year. JLARC staff also intend to provide ongoing oversight of major State expenditure forecasts.