Ministry of Social Development
and Economic Security		 Back to:
Research/Statistical Information

Forecasting the BC Benefits Caseload

Marshall McLuhan once said that humanity views the future through a rear-view mirror.

He meant that as a criticism. But to the economists who take on the monumental task of forecasting the large and complex BC Benefits caseload, the rear-view mirror is their telescope.

The annual budget of the ministry of social development and economic security is just over $2 billion. With BC Benefits making up almost 85 per cent of that amount, accurate caseload forecasting is the key to sound budgeting for the ministry. But how can anyone predict the uncountable combinations of personal, social and economic conditions that bring more than 200,000 British Columbians on and off BC Benefits each year?

The ministry’s budget for the 2000/01 fiscal year is built around a projected drop of 4.2 per cent in the BC Benefits caseload. How did the rear-view mirror direct us to that figure?

The Caseload

Programs

The BC Benefits caseload is not just one big number. There are six main programs in BC Benefits, each with its own patterns of cost and change.

Youth Works is the program for persons aged 19-24. Unless they are single parents of young children, or exempt for medical reasons, Youth Works participants are expected to be actively searching for work or upgrading their skills through one of the ministry’s training or skills development programs. The average monthly Youth Works caseload in the1999/00 fiscal year was 19,615. It has dropped by 48.7 per cent since 1995/96 (BC Benefits was introduced in December 1995).

Welfare to Work is for adults aged 25-59. Like Youth Works participants, most are expected to be looking for work or improving their skills. They also have access to the ministry’s skills programs. Average caseload for 1999/00 was 94,728, down 33.9 per cent from 1995/96.

Adults aged 60-64 are also part of the Welfare to Work program. Although they are not required to be looking for work, they are encouraged to do so. This group averaged 4,502 cases per month in 1999/00, a drop of 17.6 per cent in the last four years.

Disability Benefits Level Two consists of people who have serious, long-term physical or mental impairment that requires extensive assistance or supervision for daily living tasks, or entails unusual expenditures for transportation, special diets or other essential needs. DB2 participants are eligible for training and job search programs. Many of them wish to lead more active lives in the community, and the ministry is committed to helping remove barriers to employment for those who want to seek paid work. The DB2 caseload averaged 34,838 in 1999/00, and has grown by 54 per cent since 1995/96.

Children in the Home of a Relative are youngsters whose parents place the child in the home of a relative and are unable to pay all of the costs incurred by the relative. The CIHR caseload averaged 4,308 in 1999/00, an increase of 5.8 per cent in four years.

Seniors Benefits go to people aged 65 and over who do not receive a federal pension. They are not expected to be involved in any training or job search activity. This caseload averaged 1,812 in 1999/00, an increase of 0.9 per cent from four years ago.

Some of these programs have substantially different benefits levels and per-case costs. The basic monthly benefit for a single person in Welfare to Work, for example, is $510 ($185 living allowance plus $325 maximum shelter). For a single DB2 participant, the basic benefit is $786.42 a month.

There is a seventh program element that is part of the forecasting and budgeting process – Hardship Assistance. Hardship payments go to people who need temporary financial assistance but do not qualify for regular benefits; their assets may be too high, or they may be awaiting payments from Employment Insurance or some other income support program. Most hardship grants are repayable.

Family Size

Within each program category, the per-case costs are also affected by the number of people in the family unit.

Caseload Dynamics

The BC Benefits caseload is constantly shifting. New people apply for benefits, others leave the system because they have found a job, or enrolled in school, or found another source of financial independence. There are also changes within the caseload. People move from one program to another – from Youth Works to Disability Benefits, for example – or their family status changes. These shifts have to be taken into account, because different program and family groups show statistically different behaviour, particularly in the average length of time they are likely to need to stay on benefits.

External Factors

BC Benefits is the final economic backstop for the people of the province – the payer of last resort. Identifying, weighing and forecasting all of the many factors that could bring people to that last resort is impractical at best. One measurable factor that might seem important is the unemployment rate, but the ministry’s studies show that it doesn’t have a significant direct effect. That’s in part because it’s not a very clean statistical measure, since it’s based on monthly surveys using self-reported data on phenomena like labour force participation and job search activities.

There may also be some paradoxical effects from a falling unemployment rate. When jobs are plentiful people may be more willing to take an employment risk by dropping one job to try a new one, or by moving to another community where they think there may be more opportunity. Some of those moves will fail, at least in the short turn, and as a result some people will wind up turning to BC Benefits to help them over the crisis.

The forecast model does use unemployment rates to predict one important external event – interprovincial migration to B.C. Each month, new arrivals to the province make up about one-fifth of the people who apply for BC Benefits for the first time. The model uses unemployment predictions for B.C. and Alberta to calculate an in-migration forecast. The Alberta rate is significant because Alberta is a major source of migrants to B.C., as well as a buffer for people moving west from the rest of Canada in search of employment.

Crunching the numbers

The model builds up its forecast by making month-over-month predictions of five different components of caseload movement: Out-of-province starts, in-province starts, transfers, returning cases (people coming back on to benefits who have been off the system for less than 12 months), and stopping cases. Each component forecast requires 37 sets of calculations (called regressions), representing the combination of seven program and five family types, plus separate regressions for male and female children in the home of a relative. The transfer component requires an extra set of 37 regressions, because the system must calculate transfers into and out of each program. There are 223 separate regressions involved in running the entire forecast.

When Rob Bruce got involved in development in 1992 the ministry was using a static single-equation caseload model based on an American program. (Most American states have much less riding on a detailed, accurate welfare caseload forecast. Their programs are much simpler – some states only provide welfare for single parents, for example – and in many cases welfare is not a legal entitlement. If a state’s welfare budget for the month is fully committed, applicants are simply turned away until the next month. In British Columbia income assistance is a statutory benefit that must be provided to anyone who qualifies. Here, as in other Canadian provinces, it’s the caseload that drives the budget.)

Improvement was obviously needed, but it wasn’t easy to come by. The first attempt at writing a more sophisticated model produced a program that took more than a month of computer time to run. Today’s forecast model, known as Forecast/IA, takes about two hours to run on a Pentium 400 desktop computer, and that includes the time needed for the system to update all the data files required to run the regression analyses. Faster computers, better programming and years of development experience have made all the difference.

The hunch factor

Forecast/IA learns from experience as it sorts back through the caseload data of recent months. It uses that experience to build the month by month figures for the forecast period, which can be up to five years. However, it can’t by itself anticipate the effects of future BC Benefits program or policy changes, and it can’t always deal appropriately with the impact of changes that have just been implemented. Sometimes a program change might have a significant caseload impact in the first two or three months, then level out. Forecast/IA, however, doesn’t know that the initial effect isn’t going to continue unchanged. For this reason the model’s initial status quo forecast may have to be adjusted.

Two policy changes had to be factored into this year’s forecast. The first was the flat rate earnings exemption, which was announced in November 1999 and took effect in January 2000. The change affects singles, couples and families in Youth Works and Welfare to Work, and allows them to earn extra income ($100 a month for singles, $200 for couples and families) without it being deducted from their monthly benefits. The second change was the two per cent benefits rate increase, which was effective with the
July 26 benefits cheque.

Carefully controlled studies in the U.S. have shown that increased earnings exemptions, in particular, are likely to result in people remaining dependent on income assistance a little longer than they otherwise would. So when Forecast/IA produced a status quo forecast of a 5.2 per cent caseload decline for the year ending March 31, 2001, the hunch factor came into play. The ministry budgeted for a 4.2 per cent drop, to take into account the possible impact of the earnings exemption. Caseload performance so far this year suggests that the caution was warranted. The total caseload for the month of May was down 2.9 per cent from May 1999.

Summing up

Forecast/IA provides ministry planners with fast, detailed, easily understandable forecasts of caseload growth and decline. A companion program that is under development, Simulate/IA, will make it easier to explore "what-if" scenarios to test out the potential costs and benefits of possible improvements to BC Benefits and other ministry programs. Better budgeting and program planning allows the ministry and the government to make smarter choices in the use of tax dollars. For the 160,000 British Columbians, and their families, who make use of BC Benefits each month it will mean programs that are more clearly focused on the supports they need to move successfully from welfare to work.

A Note on Methodology

Forecast/IA generates a baseline forecast of the BC Benefits caseload. The forecast is a baseline forecast since it does not incorporate any policy changes that may be introduced after the forecast has been generated, even if the impact of the potential policy is known at runtime. The Simulate/IA module of the simulation model will generate forecasts incorporating impacts of potential policy changes.

Forecast/IA produces monthly forecasts of the BC Benefits caseload for up to five fiscal years into the future. The module is dynamic since it generates the monthly caseload forecast by predicting how the caseload will change from the previous month. The five components analyzed are new out-of-province starting cases, new in-province starting cases, returning cases, stopping cases, and transferring cases (cases moving within the BC Benefits caseload). The module relies on a series of regression models to predict how each of the five components will change over time. All the regression models are estimated at runtime using OLS. Within this structure, the module is designed to be very flexible, so that most changes to the module can be introduced relatively easily. Variables, and new regression models, can easily be added, deleted, or altered.

The current forecast is based on 223 separate regressions. For each of out-of-province starts, in-province starts, returns and stops, there are 37 regressions, representing seven programs (Under 19, Youth Works, Welfare to Work, Age 60 – 64, Seniors Benefits, Disability Benefits, Hardship) and five family types (single men, single women, couples, single parent families, two parent families) plus CIHR single men and single women. To estimate transfers, 74 regressions are required, the above 37 for transfers into a program/family type combination and 37 for transfers out of a program/family type combination. Finally, a single regression is used to estimate the flow of in-migration into British Columbia.

There are eight major steps involved in the forecast.

  1. Interpolate Unemployment Rates for British Columbia and Alberta

The British Columbia unemployment rate is used throughout the Forecast/IA module; the Alberta unemployment rate is used only to estimate the number of in-migrants to British Columbia. Forecasted values for both series are available only on an annual basis. Interpolation is used to generate monthly estimates for both series. Interpolation is accomplished by using a simple OLS model with the monthly unemployment rate as the dependent rate and eleven monthly dummy variables and the annual rate as independent variables. The forecasted data for the British Columbia unemployment rate is from the British Columbia Ministry of Finance; Alberta data is from the Alberta Treasury.

  1. Estimate In-Migration

A forecast of the number of in-migrants to British Columbia is needed to estimate the number of out-of-province new starts. A simple single equation regression model is used to estimate the number of in-migrants to British Columbia. The model is a simplified version of that used by BC Stats. The dependent variable is the number of in-migrants, while the independent variables are eleven monthly dummy variables and the average difference in unemployment rate between British Columbia and Alberta over the previous year. The use of the difference in unemployment rates between British Columbia and Alberta is that Alberta acts as both a source of in-migrants to British Columbia and a buffer between British Columbia and the rest of Canada.

  1. Estimate Out-of-Province New Starts

The number of new starts from out of province is estimated using 37 regressions. The common form of the regression models is that the number of cases is the dependent variable and the number of in-migrants in the previous three months plus selected policy-related dummy variables as independent variables. Since most in-migrants are single and employable, the models perform well for singles on Welfare to Work, Youth Works and Hardship. The Seniors Benefits, Disability Benefits, and Under 19 caseloads suffered from insufficient observations, so the mean monthly caseload for the previous twelve months is used instead of the OLS results.

  1. Estimate In-Province New Starts

The number of new in-province starts is estimated in a similar manner as out-of-province starts, again using 37 regressions. The regression models have as the dependent variable and the number of new starting in-province cases. Eleven monthly dummy variables, several policy-related dummy variables and the unemployment rate are included as independent variables. The models perform better for singles than for couples and families, better for the employable caseloads (Welfare to Work, Youth Works, Hardship) than for the less-employable caseloads (Seniors Benefits, Disability Benefits, Age 60 - 64).

  1. Estimate Transfers

Transfers are cases that are on BC Benefits in two consecutive months but change program and/or family type between the two months. Controlling for transfers is necessary because the probability of stopping differs between programs and family types. For each of the 37 program/family type groupings, the number of cases expected to transfer in to that grouping each month is estimated, as is the number of cases expected to transfer out of that grouping. Since total transfers in must equal total transfers out, transfers out are scaled so they sum to the total of transfers in.

  1. Estimate Returns

The number of cases returning each month is estimated using 37 regressions. The dependent variable is the number of cases returning each month; the independent variables are eleven monthly dummy variables, the unemployment rate, selected policy dummy variables, and the number of cases stopping IA in each of the previous twelve months. Since returns are dependent on stops, the actual number of returning cases cannot be forecasted independently of the rest of the caseload.

  1. Estimate Stops

The number of cases stopping each month is estimated using 37 regressions. The dependent variable is the number of cases stopping each month; the independent variables are eleven monthly dummy variables, the unemployment rate, selected policy dummy variables, and the number of cases starting IA (new and returning) in the previous twelve months. Like returns, stops are not independent of other components of the caseload and can therefore not be forecasted separate from returns.

  1. Estimate the Caseload

Estimates of the number of new starts and net transfers are known for the entire forecast period; however, estimates of the number of returns and stops must be calculated for each month using an iterative approach. For each month in the forecast period, the number of returns is estimated, as is the number of stops. The caseload for the month can be calculated by taking the previous month’s caseload, adding new starts, returns and net transfers in, and then subtracting stops. The data sets are then updated so that returns and stops can be estimated for the next month.

Back to: Research/Statistical Information | Ministry home

Go to top of page

Last update: August 10, 2000