labour market forecasting in australia: the science of the art

LABOUR MARKET FORECASTING IN AUSTRALIA: THE SCIENCE OF THE ARTAuthor(s): E.M. WebsterSource: Journal of the Australian Population Association, Vol. 9, No. 2 (November 1992), pp.185-205Published by: SpringerStable URL: http://www.jstor.org/stable/41110621 .

Accessed: 14/06/2014 16:51

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

Springer is collaborating with JSTOR to digitize, preserve and extend access to Journal of the AustralianPopulation Association.

http://www.jstor.org

This content downloaded from 185.44.77.146 on Sat, 14 Jun 2014 16:51:49 PMAll use subject to JSTOR Terms and Conditions

http://www.jstor.org/action/showPublisher?publisherCode=springer

http://www.jstor.org/stable/41110621?origin=JSTOR-pdf

http://www.jstor.org/page/info/about/policies/terms.jsp


Vol.9, No.2, 1992 Journal of the Australian Population Association

LABOUR MARKET FORECASTING IN AUSTRALIA: THE SCIENCE OF THE ART*

E.M. Webster Bureau of Immigration Research

PO Box 659 Caiiton South, Victoria 3053

Since 1987 nearly 50 labour market forecasts have been undertaken in Australia to assist decisions relating to government policy and budget, investment and career planning. More than 20 of these forecasts have been disaggregated by age, occupation, industry or regional labour markets. One of the chief aims of disaggregated forecasts is to help policy makers avoid future shortages or surpluses of skilled labour. A survey encompassing government departments, private research institutes and banks was undertaken to overview recent labour market forecasting exercises in Australia. This paper, which attempts to summarize these efforts, also discusses the main advantages and disadvantages of each major type of forecasting technique. Methods employed have ranged from anticipatory surveys to data-intensive input-output models.

Formal evaluation of labour market forecasts requires considerable resources and no known assessments have been conducted in Australia to date. It is unclear how significant disaggregated labour market forecasts have been in guiding the allocation of funds between competing education and training courses. Nevertheless, governments eager to avoid future shortages and surpluses of skilled labour, but less enthusiastic about forecasting, could aim to make the labour market more flexible and responsive instead. Like forecasting, however, the effectiveness of this approach has yet to be scrutinized.

'It is always difficult to forecast, especially about the future9 The Economist, 13-19 June 1992:18.

Introduction Since 1987 nearly 50 labour market forecasting projects have been

undertaken in Australia. This article, which is written for the non-economist, overviews recent attempts at labour market forecasting in Australia and summarizes contemporary views on their contribution towards enhancing the efficiency of the Australian labour market Current methods of forecasting are described and assessed purely from a theoretical perspective. The paper does not attempt to assess the accuracy or otherwise of Australian labour market forecasts as this would be a major undertaking in itself; indeed I do not know of any systematic attempt to do this to date.

Labour market forecasts, and any other forecasts for that matter, are not undertaken because it is believed that the future is predictable. Instead forecasts are undertaken because it is believed that by methodically controlling for

Thanks are due to Dick Gross» Vance Martin, Beatrice Derody, Martin Bell, Lynne Williams and the anonymous referees for comments on early drafts. Needless to say all responsibility for the content herein rests with the author.

185



major factors known to affect the forecast variable (and subsequently reducing some degree of uncertainty, eliminating forces which cancel each other out and resolving forces which are contradictory), one can systematically analyse probable outcomes of a particular course of action and plan accordingly (DEET 1991). Forecasts, then, are part of an attempt to assist decision-makers to develop long-run strategies, rather than relying on a sequence of short-run tactics.

Inevitably, decision-makers are forced to rely on some notion of a probable course of events, whether they explicitly realize it or not The use of formal forecasts is partly an attempt to make this process transparent. Whilst the repeated use of forecasts implies that forecasts of some kind are better than either random guessing or anticipating no change from the present, few forecasters or end-users in Australia systematically evaluate the value of the forecasts. This should not necessarily be counted as a criticism of forecasts, especially highly disaggregated ones. It is a major task to assess properly each round of forecasting, because of the sheer volume of resources required and the complexity of identifying whether inaccuracies were due to model failure or differences in the predicted values of exogenous variables.

Australian forecasts differ according to their regularity and whether they forecast the aggregate labour market or markets disaggregated by age, sex, occupation, industry or region. Labour market indices forecast include employment, labour force growth, wage rates and labour demand. Labour- market forecasts are predominantly used to assist decisions about government and enterprise policy: mainly education, training, immigration, welfare, superannuation, transport and town planning; budgets, in both the government and non-government sectors: tax and sales receipts, expenditures; investment; and personal career choice, especially at schools.

Functions relating to policy decisions in education, training, immigration and to a lesser extent welfare, superannuation, transport and town planning are active uses since their raison d'être is to develop further courses of action which in turn affect the future course of the labour market. Functions relating to budgets, investment and career choice imply a passive acceptance of the probable course of the labour market and only require forecasts of the labour market as inputs into the decision-maker's main focus of attention.

Naturally the assumptions adopted differ according to the end use of the forecast. Forecasts for budgetary and investment decisions are usually concerned with producing the most likely outcome, regardless of assumptions used to achieve it. On the other hand, forecasts for government planning are primarily concerned either with what will happen if present trends continue (compared with the likely outcome following some specified policy change or occurrence), or with calculating the size of measures needed to ensure a given end (see Blandy 1980).

Although more than 20 aggregate labour market forecasting projects have been undertaken for use in decisions relating to budgeting and investment, the primary focus of this paper is on disaggregated forecasts which are undertaken mainly for government policy decisions on education and training and for career guidance.

186



Within the above context, this paper addresses the following questions: what is a labour market? Why is labour-power planning important? What are the major types of forecasting projects in Australia at present? Are labour market forecasts worth undertaking? And what are the alternatives?

The Labour Market To economists, the labour market consists of two related but notionally

separate components; demand for labour, by employers, and supply of labour, from households. The labour market may be construed either in aggregate terms (the total labour market in Australia) or as separate disaggregate labour markets, occupational, industrial or regional. The choice of aggregation depends on the purpose at hand. Almost all microeconomic labour market theory is concerned with factors explaining either demand for or supply of labour, or the nature of the contractual arrangement between employers and employees.

Among the most important variables affecting the demand for labour are demand for the final product by consumers, the cost of hiring labour (predominantly the wage rate), the cost of hiring other factors needed for the production process and the chosen method of production (influenced by technology). The most important factors affecting the supply of labour by households include the cost and availability of acquiring skills needed for the job, the advantages to be had from working in a particular labour market compared with other accessible labour markets (commonly the relative wage rate) and the participation rate.

Simplified economic theory assumes that the wage rate adjusts in any given market to assure the equivalence of demand and supply so there is neither a shortage of labour nor a surplus (often exhibited as unemployment).1

Labour-power Planning A standard tenet of modern economic theory, from which dissenters must

justify any departure, is that markets will achieve the optimal societal outcome if individual members are left to pursue their own self-interest without exterior interference. A properly functioning labour market will 'allocate' labour to its highest use. However, because the labour market is conditioned by historical factors, prevailing social standards and other 'rigidities', it frequently does not 'clear', that is, achieve a concordance between demand and supply.

Major reasons for this lack of clearance include the long time it takes to educate and train a person for certain occupations, the dual role of price (the wage rate) as both a major cost of production and as a major source of income, and the large and complex nature of institutions (unions and

1 However, while this view may be sufficient for cases where the labour market is not of prime concern, it is less than satisfactory for a particular study of a labour market.

187



employer associations) in the market (see Ahamad and Blaug 1973).2 Moreover, there are large social costs in non-clearing labour markets, such as unemployment and lost production, and generally the more prolonged the non-clearance, the greater the social costs.3 Labour-power planning, which offers to reduce these costs, clearly deserves consideration as government policy.

Interest in labour market planning has been heightened since the mid- 1970s as structural unemployment (the simultaneous existence of shortages of labour in one market with surpluses in another) and 'hysteresis* have risen.4 Furthermore, an ageing labour force in Australia and most OECD countries has eroded the natural flexibility of the labour market: the ability of labour to shift easily between labour markets, which is characteristic of a youthful and growing labour force.5 It is often asserted that the failure of traditional Keynesian policies to alleviate unemployment since the mid-1970s is partly due to these high levels of structural unemployment or to hysteresis (Mitchell 1987).

The need for labour market planning does not imply the need for labour market forecasting. There are two other, less popular, approaches. One is rate of return analysis: this approach estimates and compares the pecuniary rate of return to education and training in various occupations, estimated as the discounted value of the costs of education and training and the wage differential from the base case (see Layard et al. 1971). It asserts that society would benefit from an expansion of education and training in occupations which exhibit a higher than average rate of return at the expense of occupations with a lower rate, with the absolute size of the education and training budget being determined by comparing the aggregate rate of return to education and training to the aggregate rate of return to investment in plant, equipment and new technologies. For example if plumbers had a higher rate of return than electricians then this approach would argue that society would benefit from more plumbers. The major questionable assumption upon which this approach rests, and one which probably accounts for its lack of widespread use, is the tenet that wage differentials reflect returns to education and training rather than individual abilities and artificial (legal and institutional) limits on the wage rate or demand for labour.

The second approach is the social demand approach. This view argues that the supply of education and training places should be determined by demand from households (Ahamad and Blaug 1973). Training for occupations with a higher combined pecuniary and non-pecuniary rate of return will be in

2 As product and other factor markets are becoming more complex they are increasingly the subject of economic planning too.

3 Simple examples include the significantly greater de-skilling and human misery from long- term unemployment (compared with shorter spells) and the greater disruption to production from chronic labour shortages.

4 'Hysteresis' is a form of quasi -structural unemployment caused by the de-skilling of persons who have spent prolonged spells out of the work force.

5 A faster rate of industrial change caused by factors such as technical change and the erosion of trade barriers will put even more pressure on the labour market to respond.

188



greatest demand, and society will benefit from an expansion of these occupations so long as their rate of return is above the norm. This approach embodies similar reasoning to the 'rate of return* approach but implies that prospective students and trainees have sovereign and informed choice rather than the distribution of student places across courses being determined by a third party, the government. Inherent in this reasoning is the necessity for students and trainees to be able to control the distribution of resources between courses, either by paying fees themselves or by using educational vouchers.6

While this approach suffers from the same weaknesses as the 'rate of return" approach, it also relies on students and trainees having, at the margin, a reasonable perception of future relative wage rates and employment opportunities. Ironically, it is for this purpose that schools and vocational guidance agencies often request labour market forecasts.

To summarize, while labour market forecasting is not the only type of labour-power planning technique, it approaches the problem of labour-power planning from a perspective which embodies slightly different simplifying or "as if assumptions than the alternative rate of return and social demand approaches. If sizeable resources were assigned to labour-power planning as a whole, forecasting would be a complement to, rather than a substitute for, these other approaches.

Types of Labour Market Forecasting in Australia There are several ways to categorize forecasting: by method used, by end-

user type (policy, budget, investment, career choice), by level of aggregation (aggregated, disaggregated) and by forecast horizon (short, medium and long-term). The forecasts here will be distinguished according to the method employed. The approaches described are not mutually exclusive, though some lend themselves more easily to forecasting the supply side of the market (e.g. the cohort-component method) and some to the demand side (e.g. the lead indicator). Some approaches, in particular the large disaggregated econometric models, incorporate a number of the other methods listed.

The advantages and disadvantages of each method are discussed in each section to indicate which method is appropriate for a particular use and to highlight the limitations which should be borne in mind when interpreting results. Shortcomings of forecasting methods may not imply that a particular method is inappropriate as the cost of using a more complex approach may outweigh the benefits.

The methods are categorized as follows: a. non-statistical methods

i. expert opinion ii. anticipatory survey iii. scenario writing

6 These are coupons granted to a person by a benevolent party (e.g. government) and only redeemable on certain expenditures, in this case education.

189



b . statistical methods i . extrapolation (trend projection) ii. autoregressive models (AR, ARMA and ARIMA) iii. lead indicator iv. density ratio v. cohort-component vi. comparative country vii. single econometric equation viii. general economy-wide models

The following sections provide examples of Australian forecasts employing each technique; classification is based on the forecasters' assessment of their own method. Further details of the characteristics of these forecasts are set out in Tables 1 to 4.

Non-statistical methods The main advantage of non-statistical methods is that they allow for the

incorporation of information which either is not routinely measured in standard data collection (such as expectations and technological develop- ments), or cannot be collected by normal methods in a timely fashion. The latter relates especially to short-term forecasts of the economic trade cycle.

Ironically, while one of the stated reasons for formal forecasting is to make clear and explicit the values one is attaching to future variables, non-statistical methods inevitably assume explanatory variables which are implicit and cannot be assessed and evaluated. While this problem can achieve massive proportions when large disaggregated long-term forecasts are attempted, and accordingly this method is generally unsuitable for this purpose, non- statistical approaches have been used in conjunction with both fixed coefficient and economy-wide models (e.g. Australian Department of Employment, Education and Training (DEET), see Table 3).

Expert opinion This is used by the Australian Treasury, NSW Treasury and WA

Treasury, see Table 1. Essentially this method, of which one example is the 'Delphi' method, is

used when a small group of persons, considered to have expertise in the area, assemble to pose their views on the future and reach consensus on a likely scenario. It is commonly used to produce short-term macroeconomic forecasts of aggregate labour market variables. For example, forward estimates of aggregate employment and unemployment are made to feed into forecasts of government revenues and expenditures (e.g. taxes and welfare benefits) and trade accounts. Similarly, future movements in the Consumer Price Index (CPI) and aggregate wage growth are estimated to adjust price variables in budget forecasts.

190



Table 1 Characteristics of labour market forecasts using non-statistical methods. Australia, 1987 to 1992 (June)

Method Region Forecttt Level of Forecast Forecast Uses Variable Dit aggregai i on Horizon Frequency

i Expert opinion Aust Treasury Aust EPU none short annual budget NSW Treasury NSW EPU none short annual budget SA Treasury S' EPU none short/med quarterly budget WA Treasury WA EPU none short quarterly budget

ii Anticipatory s'vey AustDEET Aust DB occ short annual gen/ed/other Latrobe Reg Comm Vic ESU ind/reg long annual other WA Dept EXTAFE WA B occ short annual gen/ed ACM Aust, Vic EPU none med/long quarterly gen

Where: Disaggregation refers to type of disaggregation on the labour market variables. E=employmentt P=population, U=unemployment, D=demand for labour, S=supply of labour, B=market balance, occ=occupation, ind=industry, reg=region. gen=general public information, ed=education and/or training planning, oth=other uses. Forecast Horizon: short = 1-2 years, medium- 3-4 years, long = 5 + years.

Source: Author's Telephone & Mail Survey 1992.

Anticipatory survey This method is used by the Australian Department of Employment,

Education and Training (DEET), Latrobe Regional Commission, WA Department of Employment, Training and TAFE, and Australian Chamber of Manufacturers (ACM) (see Table 1).

Survey forecasts, frequently disaggregated by occupation or industry, are usually provided as general information for businesses, training bodies, employment programs and school career advisory centres. Their principal purpose is to predict the demand for labour or employment. They commonly take the form of telephone or mail surveys of employers asking about their expectations as to future business prospects, and may include estimates of labour demand and labour shortage.

The main problem with surveys which query the labour market is that labour-market decisions tend to be lag rather than lea d indicators of economic activity. Thus it has been suggested that it would be more appropriate to survey anticipations of investment, sales or housing starts and then use these as inputs into a model which relates these variables to the labour market, than to try to survey anticipations of the labour market directly. Surveys of investment expectations are generally considered the most reliable (Granger 1989: 155).

Another disadvantage, common to many areas of economics, is that the sum of the parts does not equal the whole: if everyone acts on the basis of their expectations then the expected outcome will not totally materialize. Furthermore, employers ordinarily do not have a view of the future which differs substantially from their present experience. Anticipatory surveys do, however, have the advantage, subject to survey size, of giving highly disaggregated labour market data.

191



Scenario writing (no Australian examples discovered) These are generally qualitative forecasts (although often incorporating

some data analysis) which attempt to paint a picture of the future, the economy or life in general. The most infamous examples include The Limits to Growth (Club of Rome 1972) and Future Shock (Toffler 1970) with recent examples including The Market Experience (Lane 1991) and Millennium (Attali 1991). They are interdisciplinary and are based on a mixture of specialist knowledge and the practical wisdom of long experience. A common aim of such scenarios is to promote discussion and give early warning of looming catastrophes (Attali 1991) or to describe a Utopia to be aimed for (Lane 1991; see Blandy 1980).

Statistical methods The eight statistical methods outlined below rely primarily on a long series

of historical data. An exception is comparative country methods. While each has specific limitations, these methods collectively suffer from two weaknesses, save when the forecaster has been diligent enough to account for them.

They frequently fail to take into account how labour market conditions at the time of data collection have affected the data (ASTEC 1983). A partial solution to this problem is to state that the forecasts represent an increase or decrease in the labour market variable relative to what it was in the base year (DEET and Victorian Department of Business and Employment in Table 3). However, this begs the question as to the precise state of the labour market in the base year and how to correct for labour market imbalances of the observed magnitude. Most assume that relative wages, the ratio of labour of differing skills within the industry and firm, the extent of labour hoarding7 and the extent of labour shortages or surpluses remain constant into the future (Wabe 1974,Layardetal. 1971).8

Extrapolation (trend projection) This is used by ACT Chief Minister's Department, Australian Bureau of

Statistics (ABS), Centre for International Economics (CIE), Australian Department of Employment, Education and Training (DEET), EPM Consult- ing Group, Queensland Department of Employment, Vocational Education, Training and Industrial Relations (EVETIR), WA Department of Planning and Urban Development, ANZ Bank, Commonwealth Bank, Macquarie Bank; see Table 2.

7 Labour hoarding is the extent to which employers employ labour over and above current production requirements. Usually this is done to avoid losing specifically skilled labour necessary to meet demand when business improves.

8 While there is some evidence to suggest that relative demand for labour is insensitive to relative wages (although this depends on the level of disaggregation), this may be observed because of rapid shifts in labour supply or because employers, for other reasons, offer non- wage inducements to attract labour.

192



Table 2 Characteristics of labour market forecasts for discrete markets using statistical methods August 1987 to 1990 (June)

Method Region Forecast Level of Forecast Forecast Use» Variable Disaggregation Horizon Frequency

i Extrapolation ACT Chief Min Dept ACT W Age/sex/ind/reg medium irregular gen/ed/other ABS Aust SP age/sex long not known general Œ Auft S age/sex long not known other Aust DEET Aust EPST age long not known gen/ed EPM Consult group Aust SE age/occ/ind long irregular general Qld Dept EVE™ Qkl A trade short irregular education Qìd Dcpt E VETIR Qid ST age/sex long irregular education WADeptP&UD WA S age/sex long biennial transport ANZBank Aust EPU none short/med monthly investment C'wealth Bank Aust HJ none short monthly investment Macquarie bank Aust EPU none short monthly investment

ii Autoregressive Bankers* Trust Aust EPU none short quarterly investment

iv Density ratio Ratio Consultants Vic E ind/region long irregular transport Qld Dept of Transp QU E region long irregular transport

v Cohort component see Table 3 below vii Single equation Vic Dept of Educati Vic A trade short annual ed Vic Dept of BAE Vk PS age/sex/reg long biennial gen/ed/other

Where: Disaggregation refers to type of disaggregation on the labour market variables. E=employment, P=population, U=unemployment, S=supply of labour, T=student enrolments, A=apprentice commencements, occ=occupation> ind=industry, reg=region. gen=general public information, ed=education and/or training planning, oth=other uses. Forecast horizon: short = 1-2 years, medium = 3-4 years, long = 5 + years.


This most simple and popular form of statistical forecast is inevitably the starting (or finishing) point for most forecasting exercises. As such its uses cover the entire spectrum from general public information, through use in enabling enterprises to estimate demand for their output, to estimates of housing starts for banks. Both aggregated aiid disaggregated forecasts have been produced using this method. Extrapolation involves fitting a trend line to a historical data series of the variables to be forecast which, in practice, are usually demand for labour (proxied by employment) and labour supply. Underlying functional forms include:

Et = a + bt linear trend

Et = ea+bt exponential curve

Et = a + br* modified exponential curve

Et = a + bt + ct2 parabolic curve where: E is the forecast variable

a, bande are coefficients t represents time e is the natural number.

193



While the main attraction of this method is its simplicity, with respect to ease of comprehension and minimal data needs, its major drawback is its 'black box' nature which does not allow the user to understand the economic and social processes which have generated the underlying trend. For some purposes, however, the key requirement may be for a forecast that is cheap and timely to produce, and the 'black box' aspect of the model may not be of practical import

Trend analysis has the drawback that it cannot pick crucial turning points in the forecast variable. According to Curtain (1989), the effects of technological change are erratic over time and using trend analysis to predict these effects may produce ridiculous results. He suggests that a case study approach to estimating the effects of technological change is required (see also ASTEC 1983). However, simple extrapolation is useful in detecting mutually inconsistent trends, such as rates of growth in the labour force and the participation rate (Blandy 1980).

Autoregressive models (AR, ARMA andARIMA) (e.g. Bankers' Trust, see Table 2)

Banks use aggregate employment and unemployment forecasts to predict housing starts (and bankruptcies) and as factors which can predict the profit- ability of businesses and the stock market Some banks, especially merchant banks, also supply forecasts as part of general customer service. Like basic extrapolation methods, autoregressive models arc thinly veiled 'black boxes' which rely on estimating the relationship between the forecast variable and its previous values, called a lagged dependent variable. Hence, the forecaster is concerned with estimating a regular relationship with the change in the forecast variable, rather than its absolute magnitude (as is the case with trend projections).9

A simple autoregressive curve (AR(1)) is:

Et = aE^ + ut such that -l<a<l (to ensure a non-explosive model)10 where u represents a classic error term.

Where the forecaster believes that the components of the forecast variable obey the simple autoregressive model, or obey a combination of simple autoregressive and moving average models, then the autoregressive-moving average (ARMA) model is more appropriate. An example of a simple ARMA(1,1) model is:

Et = a¡Et_i + ut - but_i .

More complex integrated mixed autoregressive-moving average (ARIMA) models are applied when a higher level of autocorrelation is present in the forecast variable's historic data. Unlike trend projections, autoregressive

9 More complex vector autoregressive (VAR) models may also be used. VAR models regress the dependent variable not only on its own lagged values but also on the current and lagged valued of a selection of independent variables.

10 If a=l, the process is integrated, e.g. ARIMA. Also, this case mimics the trend models in Extrapolation above.

194



models of orders greater than or equal to two can predict turning points in the forecast variable and their modest data requirements make them suitable for many purposes.

Lead indicator There are no Australian examples where lead indicators are used on their

own but this technique is integrated into most demand-side forecasts. In an attempt to overcome the difficulty of predicting crucial turning

points, lead indicators are sometimes combined with other methods. The desired forecast variable is based on some known variable which is believed to precede the forecast variable by a regular time interval. Choice of lead indicators is based on a combination of a priori reasoning and intensive examination of the cyclical relationships between a broad set of economic variables. Thus lead indicator techniques could be described as partial 'black boxes'. Lead indicators may be used in either aggregated or disaggregated forecasts.

This method is suitable only for very short-term forecasts, as the few lead indicators which exist do not precede the labour market by longer than perhaps 12 to 18 months. Lead indicators are often implicit in other forecast methods and are generally favoured by journalists and others who wish to be seen as the first to detect a turning point. The stock market is often cited as a lead indicator of major economic variables (reflecting to some degree business expectations) but, according to Granger, its record is poor (Granger 1989: 167-176).11 More reliable lead indicators include physical capital utilization (cycle peak only), inventory investment, investment commitments, residential production, job vacancies and overtime worked (cycle peak only). The first three indicators are leads for production from which (with a lag) the labour market can be forecast.

A further problem with lead indicators is that they can produce false signals and for this reason forecasters usually prefer to refer to a collation or index of indicators before they pronounce the arrival of a turning point

Density ratio

Examples include Ratio Consultants, Queensland Department of Transport, see Table 2.

The density ratio method relies on identifying an optimal or historically constant ratio between the forecast variable, usually demand for labour, and a physical measure of output, for example students per teacher, population by age cohort per medical practitioner (Ahamad and Blaug 1973). It is only suitable for selected occupations, industries or regions for which a stable or optimal relationship can be found and for which data are available. The Queensland Department of Transport assumes on a disaggregated regional

11 Perhaps this reflects the lack of perspective which private and institutional equity investors have over the whole economy or the influence of other (speculative) forces on the stock market.

195



Method Region Forecast Level of Forecast Forecast Uses Variable Disaggregation Horizon Frequency

Table 3 Characteristics of labour maiket forecasts for the whole economy using statistical methods Australia, 1987 to 1992 (June)

Disaggregated forecasts CSAES SA E age/sex long irregular gen/ed/other AustDEET Aust EPST age/sex/occ/ind long not known gen/ed NlhlK Aust ESDI) age/sex/occ/ind medAong annual general Public Information Vic E occ/ind médium irregular education Systems Vic Dept of B<fcE Vic DSB occ/ind/qualif i - long annual gen/cd/cg/

cation other Vic Dept of B&E Vic ES trade medium biennial gen/ed/cg/

other

Aggregated forecasts Access Economics Aust, Vic EPU none medAong quarterly general Aust Treasury Aust EPU none not known not known not known COPS Aust EUS none short biannual general National Aust Bank Aust EPU none medium monthly investment Qld Treasury QW HJW none short/med biannual budget Syntec Eco Services Aust EPU none medium quarterly general Syntec Eco Services Aust E none medium biannual general Tas Treasury Tas EPU none short quarterly budget Vic Treasury Vic EPU none short annual budget Vic Treasury Vic EPU none long biennial budget/other Westpac Bank Aust EPUS none short annual investment

Where: Disaggregation refers to type of disaggregation on the labour market variables. E=employment, P=population, U=unemployment, D=demand for labour, S=supply of labour, B=market balance, W=wages, occ=occupation, ind=industry, qual=qualification, gen=general public information, ed=education and/or training planning, cg=careers guidance, oth=other uses. Forecast horizon: short= 1-2 years, medium= 3-4 years, long= 5+ years


level that the employmentrpopulation ratio is fixed and uses population forecasts to predict regional employment

The density ratio method relies on the forecaster either presenting a range of forecasts based on variations in the density ratio or on .making an assumption about what the future ratio will or should be (even if the forecaster states that it is constant). Furthermore it requires the forecaster to make forward estimates of the explanatory variable: in the examples given above, students or the population by age cohort.

Cohort-component Examples are Victorian Department of Business and Employment, and

most forecasts which have a disaggregated supply side; see Table 3. Many of the single-market methods listed above concentrate on estimating

the demand side of the market, or employment, because when there is an excess of labour supply over demand, it is the short side of the market, demand, which determines employment. The cohort-component method is almost exclusively used for predicting the supply of labour, either in

196



aggregate or by occupation, sex or age. All but one of the Australian examples listed in Table 3 use the cohort-component method as the supply side of their economy- wide models.

The cohort-component method takes cohorts of persons by age, occupation, or sex at a particular time and projects them by subtracting exits from the cohorts (e.g. through death, occupational wastage) and adding new entrants (e.g. persons aged 15 years, new education completions).

One major drawback to this method is that it generally assumes constant exit and entrance rates over time: the validity of which depends on the use, level of disaggregation and particular labour market being forecast Moreover, when applied to occupational matching, it frequently does not take into account retraining and re-education, thus tending to overstate shortages or surpluses of labour. Like most methods reviewed in this article, it assumes an unvarying relationship between qualifications and occupation or industry. While few forecasters believe that this is a highly probable outcome, it is consistent with the notion of only varying those parameters which can reason- ably be gauged.

Comparative country (no Australian examples discovered) The comparative-country method uses the profile of a foreign country

considered with respect to the forecast variable to be more developed' than the domestic country to identify where the latter will be in the future. While it is a popular method in newly industrialized and less developed countries, it can also be used in Australia to the extent that its social norms, techniques of production and forms of work organization resemble those in earlier years in North America and northern Europe. Classic examples are the female participation rate, the preponderance of part-time work and the education and training rates of youth. Similarly there is a lagged but imperfect relationship between overseas trade cycles and those of Australia.

The obvious problem with this method is that there is no certain way of estimating how long it will take country 'a' to catch up to country 4b' or indeed, whether it will follow the same path at all. Any estimates will naturally set a limit on the forecast horizon. For example, if it is assumed that the USA is '18 months ahead' of Australia, then only an 18-month forecast can be produced using this method. For this reason forecasters will often use overseas standards as maximum rates which may be approached by a mono- tonic or autoregressive trend line forecast.

Single econometric equation This is used by Victorian Department of Education, Victorian Department

of Business and Employment; see Table 2. The Victorian Department of Education has used a single econometric

equation to predict the demand for and supply of apprentices by trade. Estimation of single equations is often used when a single labour market (either the whole labour market or a single occupation, industry, demographic cohort or region) is being forecast. Forecasters use regression techniques to estimate the independent effects of explanatory variables, such as production,

197



technical change, relative wages and demographic factors, on the variable to be forecast, such as employment, demand for labour, supply of labour or wages.

A frequently used functional form for the demand for labour is the Cobb- Douglas:

Et = AQt1/aK;g/ae^/a

where: E is employment or demand for labour Q is production K is physical capital e is the natural number t represents time a is the elasticity of substitution of employment with respect to production

g is the elasticity of substitution of physical capital with respect to production

d captures autonomous growth in production over time (basically due to improvements in capital and management technologies)

A represents a constant. As with all estimates based on regression analysis, before they are considered acceptable, estimated parameters must meet specific statistical conditions, such as stability, consistency and unbiasedness.

Estimating this equation is not always possible given the lack of suitable data in many labour markets, especially for production and physical capital. In general the more adaptations and transformations of the original data that are required to meet the specification of the equations, the greater the problems in using the estimated coefficients to make forecasts. This is especially problematic when one is required to transform the forecast variable, and leaves some doubt as to whether one is forecasting the desired concept or the published statistic.

A further problem with single-equation models is that they implicitly assume that the outcome of the labour market in question will not affect other sectors of the economy to such an extent that this feeds baók to the original labour market. While this is probably a reasonable assumption for small fish in a big sea, it is not reasonable if the labour market employs a large enough portion of the population to affect aggregate income in die region at hand, or if the industry is an upstream activity for other sectors (e.g. critical component supplier).

By the insertion of plausible assumptions for the explanatory variables in the estimated equation, conditional forecasts are obtained, for example that employment will be V in year V if production is 4z' To obtain unconditional forecasts, it is necessary to make independent forecasts for each of these explanatory variables and simple autoregressive models of these explanatory variables are often used.

This widespread need for forecasts of the explanatory variables and the general interdependence of most economic relationships give rise to the preference for general economy- wide models.

198



Table 4 Characteristics of labour market forecasts using unknown or variable statistical methods Australia, 1987 to 1992 (June)

Method Region Forecast Level of Forecast Forecast Uses Variable Disaggregation Horizon Frequency

Unknown Kinhill Engineers Vic E sex/ind/reg long irregular transport Bis Shrapnel Aust EPUS none long annual general Bis Shrapnel Aast EFUS none short monthly general Varíes VicDeptofB&E Vic DS occ/qual long varies gen/ed/other NSWDeptlR&E NSW varies varies long varies gen/ed/other

Where: Variable refers to the forecast variable, Disaggregation refers to type of disaggregation on the labour market variables, Period refers to the forecast horizon. E=employment, P=population, U=unemployment, D=demand for labour, S=supply of labour, occ=occupation, ind=industry, qual=qualification, reg=region. gen-general public information, ed=education and/or training planning, oth=other uses. Forecast horizon: short= 1-2 years, mediums 3-4 years, long= 5+ years


General economy-wide models

Examples are Centre for South Australian Economic Studies (CSAES), Australian Department of Employment, Education and Training (DEET), National Institute of Economic and Industry Research (NIEIR), Public Information Systems, Victorian Department of Business and Employment, Access Economics, Australian Treasury, Centre of Policy Studies (COPS), Queensland Treasury, Syntec Economic Services, Tasmanian Treasury, Victorian Treasury, Westpac Bank; see Table 3.

Given their size and complexity, these models are designed for many forecast-related uses. Only three (CSAES, DEET and Victorian Department of Business and Employment) concentrate primarily on the labour market The uses of the labour market components vary from public information and government labour market program planning to budget and investment planning.

Estimation of single equation forecasts can often be misleading if the effects from the labour market in question to elsewhere in the economy feed back to this labour market causing the original assumptions to be invalid As a simple example, a fall in supply of a particular type of labour may lead to an increase in its wage rate, a rise in the supply and employment of substitute types of labour and a change in workplace technology, all leading to a fall in demand for the type of labour in question (see Ahamad and Blaug 1973).

To overcome some of these problems, economists resort to general models of the economy (basically a series of simultaneous equations) which incorporate interconnected sets of equations for each market in the economy (see Ahamad and Blaug 1973; Bosworth, Evans and Lindley 1974; Burgan 1989). Thus, interdependencies between different markets can be taken into account. The effects of a given change in one market, such as a rise in the world price of oil, can then be traced through other sectors of the economy.

199



In Australia, there are at least nine general models which forecast the aggregate labour market, of which at least three potentially allow for the production of large numbers of disaggregated labour market forecasts. While these models differ with respect to the specific methods used to forecast output, particularly by industry, they all essentially employ the same technique of deriving forecasts of employment by industry and occupation.12

In essence, model builders derive input-output tables comprising industry output by employment and industry employment by occupation; these are drawn from surveys and censuses conducted by the Australian Bureau of Statistics (ABS). If required, separate matrices can be derived for regions. These matrices are then applied to forecasts of industry output to estimate future demand for certain occupations, industries and regions. In single equation notation this may be depicted as:

Ei = X(Eij^j)(Ej/Qj)(Qj) such that

EEi/Ei-l where: i represents an occupation

j represents an industry. Further disaggregation into markets for specific qualifications can be

achieved by applying occupation by qualification matrices to the employment by occupation and industry matrices:

E^KEfc/EOEj where k represents either education or qualifications.

Depending on the starting point, this method can be used to derive future estimates of employment or labour demand.

To realize a forecast of the whole labour market, and thus obtain estimates of impending future labour shortages and surpluses, a companion model of labour supply must be generated, the most popular method being the cohort- component method, in conjunction with assumptions about education and training levels.

While meeting some of the deficiencies of the methods described earlier, these general models are not without their own inadequacies. They frequently assume that labour productivity (Ej/Qj), the occupational structure (Ey/Ej), qualifications structure (E^/Ej) and (where relevant) inter-regional labour mobility are fixed at the point of time when the matrices were recorded. This is a particular problem in sectors of the economy where the progression towards multi-skilling is changing not only occupational definitions over time, but also the relationships between occupations and industries, occupations and qualifications and between different occupations (Layard et al. 1971).

12 One of the forerunners of this approach was the fixed coefficient model developed in the early 1960s, called the Mediterranean Requirements Project or MRP. The fixed coefficients and general-equilibrium models are the only forecasting methods which seek to forecast simultaneously a multitude (perhaps over 500) of disaggregated labour markets.

200



These simplified assumptions aie usually adopted in the absence of better, and cheaply obtained, information. Where they are perceived to be misleading, estimates can be made using any of the methods described above, including the non-statistical methods, to build in varying rates and structures of labour productivity, occupational structure, qualifications structure and inter-regional mobility (see Bosworth et al. 1974; DEET 1991). For example labour productivity is commonly assumed to follow a positive trend increase in each industry, using simple trend extrapolation; see extrapolation (trend projection).

Other shortcomings of these general models relate to the procedures used for forecasting output (by industry) and as this subject is very extensive it is not pursued here. Major problems include assumptions of fixed coefficients and equation mis-specification (see for instance Wabe 1974; Granger 1989).

All models need to have a way of estimating lead variables and the following techniques may be engaged here: expert opinion, anticipatory survey, extrapolation, autoregressive models, lead indicator and comparative country.

Assessment of Labour Market Forecasting Ultimately the assessment of a forecast is a cost-benefit exercise and must

depend on its ability to predict more accurately than the next-best alternative: usually either random guesses or assuming no change from the present. Theoretically this involves assessing many unobserved variables such as the alternative, non-forecast predictor which decision-makers would have used, the weight given to forecasts in decision-making, the often unmeasured costs of producing forecasts, and the economic consequences of making decisions based on the alternative predictor. However, this is a nearly impossible task. As a proxy, formal assessments of forecasts have used measures such as the deviation of the forecast variable from the realized value to rank forecasting methods, including a no-forecast option.

No formal assessments appear to have been undertaken of any disaggregated Australian labour market forecasts. Consequently little comparison can be made either between different sets of forecasts or with a 'no-forecast* alternative. Rankings of various overseas forecasting models, including the no-forecast option, have concluded that past forecast success does not provide a reliable guide to future performance. Wallis (1987) found when ranking several large UK economic models that there was little consistency when models were ranked either by forecast variable or by different 'rounds' of forecasting. It is not uncommon for forecasts to be right for the wrong reason: occasionally when the forecast affects people's behaviour it may invalidate itself. Ahamad and Blaug (1973) deem a forecasting method 'worthy' only if past errors can be identified and applied to reduce further errors.

According to Granger (1989:192) it is difficult to derive immutable laws on the best way to specify and estimate econometric equations for providing forecasts. An equation or model which may be superior to others today, may perform poorly in future years. A good fit (ex post) does not always imply

201



good (ex ante) forecasts, short-term models do not necessarily predict as well as long-term ones, models with adaptive parameters have not done better than models with fixed parameters and complexity does not always improve accuracy.13

Despite this, Granger (1989:192) has specified four general rules for forecasting: the further ahead the forecast the greater the error, the greater the (relevant) information set the smaller the error, the more data analysis the smaller the error, and a forecast that is the average of more than one different approach is usually better than its components.

The usefulness of a forecast depends on the user knowing its limitations: in particular the limiting assumptions adopted to produce the forecast and how these might affect a particular application. While highly disaggregated forecasts are often viewed as being of most use in areas such as education, training and immigration planning, they also carry greater inbuilt sources of error, making their value over less disaggregated forecasts one of judgement rather than fact. For example, the greater the level of disaggregation the greater the possibilities for substitution between labour and capital and between labour of different types; the greater the changes in job design, output type, work organization and technique of production; and the more sensitive the effect of social changes such as on the supply of labour to selected labour markets. One analysis of disaggregated USA employment forecasts between 1976 and 1982 calculated a mean absolute error of 20 per cent on each forecast variable (Goldstein and Cruz 1987).

A useful exercise, but one which is beyond the scope of this paper, would be to estimate the weight given to forecasts in decision-making in Australia. Anecdotal evidence on the use of forecasts in allocating funds between and within institutions of higher education suggests that their role has been limited. While it appears that historical allocations and general bargaining between faculties dominate the allocative processes, some role is given to "student demand" for courses, which is more akin to the social demand approach described earlier than to forecasting. Reports indicate that the role of forecasts in determining allocation of funds in technical and further education is mixed and may vary between states.

It is possible that education and training committees are composed of persons who see their role as representing their sectional interests, enhancing education and training standards or protecting 'empires', rather than undertaking an obligation to balance adequately a future labour market. There is cause for concern in the possibility of a barrier between the producers and consumers of disaggregated labour market forecasts in the education and training area.

13 Granger (1989) believes that there is support for the view that econometric models do not necessarily produce better forecasts than un i variate time series projections. Two large studies examining the relative performance of large econometric models over more simple trend extrapolations have yielded contradictory results. Ash, Smyth and Heravi (1990), in an assessment of over 7000 OECD forecasts, concluded that forecasts from econometric models were superior, while Ashley (1988) in an assessment of short-term macroeconomic USA models found trend extrapolations to be superior.

202



Over die past decade labour-power forecasting has lost favour as a method of avoiding labour-market surpluses and shortages, particularly overseas (Psacharopoulos 1991). Many of the overseas criticisms of forecasting are, however, based on a fairly mechanistic view of its application, which is not the case with more sophisticated users in Australia; for example Psacharopoulos (1991) in his critique of forecasting assumes that forecasts are used for "head counting9 styles of policies and decisions which exclude the costs of education and training.

Alternatives to Labour-market Forecasting (or Planning) Decision-makers rely upon non-forecasting tools for achieving their goals

either because of a view that labour market forecasts, especially disaggregated forecasts, are too unreliable or because they believe that there are diminishing returns to forecasting, such that only a small number of forecasting exercises warrant their cost (Blandy 1980; ASTEC 1983). While there are few practical substitutes for forecasting in the area of budgets and investment, alternative approaches to labour market analysis can be found to guide government policies aimed at reducing structural mismatch across labour markets. Policy- makers may aim to make the labour markets more flexible and responsive so they respond Automatically' without government direction to ensure that shortages and surpluses of labour do not persist

While even the most enthusiastic labour market reformer does not believe that a fully informed and flexible labour market will ever be achieved, adherents of this view believe that the gains per unit cost from making the labour market more flexible and responsive outweigh the gains per unit cost from pursuing policies based on labour market forecasts.

Examples of mechanisms to enhance labour market flexibility include (Kirby 1985): • providing more accurate and more timely contemporaneous information on

labour markets to participants; • ensuring young people gain broadly based transferable skills so they can adapt easily to varying occupations and industries in later life. At present there are too few training opportunities for young people who have below average educational achievements;

• ensuring that the quality of education and training courses are fully articulated and responsive to employer's requirements (note that this mechanism involves a trade-off with the previous mechanism unless total years in initial education/training increases);

• increasing mobility between occupations by eroding artificial licensing barriers, quantity limitations on bridging courses and workplace demarcation rules. This includes ensuring uniformity and transparency of vocational skills.14 The latter will increase the portability of skills between industries and States and subsequently raise the individual's desire to invest in these skills;

14 Transparency implies a widely accepted certification system so sets of skills from differing sources can be compared.

203



• encouraging labour force participation by mothers who may provide scarce skills, by increasing the flexibility of working times, shopping times, child care, and so on;

• increasing adult education and training opportunities; • providing positive assistance to persons displaced by technological change

and government policy change; • freeing relative wage rates from non-economic forces, thus enhancing their

ability to act as signals of relative employment opportunity; and • most importantly (as this is the target of much labour market forecasting)

making the quantity of education and training places responsive to demand by prospective students (subject to academic prerequisites). Some of these factors are currently being addressed in Australia as part of

policies aimed towards broader objectives. Examples include award restruc- turing, changes within TAFE to both accredited and non-accredited courses and schemes to encourage the return of mothers to the work force.

On the last point, Psacharopoulos (1991) recommends the use of longi- tudinal surveys of ex-students and trainees to determine the 'value' of their education and whether it should be expanded or not The OECD (1990) anticipates that during the 1990s, governments will increasingly assist workers before they are laid off and businesses will aim to increase their workers' ability to cope with change rather than seek to replace their incumbent labour force from the external labour market.

While many of these policies are not new in labour economics, there has been no examination of their efficiency (value per cost) as alternatives to labour market forecasting in reducing shortages and surpluses of skilled labour.

Conclusion In this paper, labour market forecasting in Australia has been examined

from a conceptual perspective. A survey of statistical and non-statistical methods has been made. It has not been possible to assess technically the value of labour market forecasts. The existence of regular buyers of forecasts, especially aggregated ones, is 'proof in standard economic terms of their value. This is less frequently the case for disaggregated forecasts which are largely produced by governments for general information and for education and training institutions.

Whether forecasts are used for the purpose for which they are intended is not clear and should be more thoroughly scrutinized. Mechanisms which enhance labour market flexibility should reduce the problems which arise from labour market mismatching, thereby reducing to some extent the need for, and problems associated with, labour market forecasting. However, the relative costs and benefits of these policies are also unmeasured. Therefore when choosing between the options of forecasting and labour market flexibility, one is forced to rely on informed judgement

204



References Ahamad, Bashir and Mark Blaug. 1973. The Practice of Manpower Forecasting. London: Elsevier

Scientific Publishing Company. Ash, J.C.K., DJ. Smyth and S.M. Heravi. 1990. The accuracy of OECD forecasts of the

international economy: demand output and prices. International Journal of Forecasting 6(3):379-400.

Ashley. R. 1988. On the relative worth of recent macroeconomic forecasts. International Journal of Forecasting 4(3):363-376.

Attah, Jacques. 1991. Millennium. London: limes Books. Australian Department of Employment, Education and Training (DEET). 1991. Australia's

Workforce in the Year 2001. Canberra: Australian Government Publishing Service. Australian Science and Technology Council (ASTEC). 1983. Technological Change and

Employment: A Report to the Prime Minister. Canberra: Australian Government Publishing Service.

Blandy, RJ. 1980. Report of the South Australian Tripartite Working Party on Manpower Forecasting. Adelaide: South Australian Department of Industrial Affairs and Employment.

Bosworth, D.L., GJ. Evans and R.M. Lindley. 1974. Mechanistic manpower models. In J. Stuart Wabe (ed), Problems in Manpower Forecasting. Westmead: Saxon House.

Burgan, Barry. 1989. Employment requirements, forecasts for the South Australian economy, 1990-2005, Stage I- Evaluation of forecasting methodologies. Adelaide: The Centre for South Australian Economic Studies (mimeograph).

Club of Rome. 1972. The Limits to Growth. London: Pan. Curtain, Richard. 1989. Addressing skill shortages: the limitations of macro-level labour market

forecasting and an alternative approach. In Australian Skill Shortages, Papers from the conference on Australian Skills Shortages and Labour Market Data held by the Centre for Research in Education and Work (CREW) School of Education, Macquarie University, 1988.

Goldstein, H.A. and A.M. Cruze. 1987. An evaluation of state projections of industry and occupational employment. Monthly Labour Review 110(10):29-38.

Granger, C.W.J. 1989. Forecasting in Business and Economics. San Diego: Harcourt Brace Jovanovich.

Kirby, P.E.F. 1985. Report of the Committee into Labour Market Programs. Canberra: Australian Government Publishing Service.

Lane, Robert E. 1991. The Market Experience. Cambridge: Cambridge University Press. Layard. P.R.G., D. Sargan, M.A. Ager and DJ. Jones. 1971. Qualified Manpower and Economic

Performance: An Interplant Study of the Electrical Engineering Industry. London: Allen Lane Penguin Press.

Mitchell, W.F. 1987. The NAIRU, structural imbalance and the macroequilibrium unemployment rate. Australian Economic Papers, 26(48).

Organization for Economic Co-operation and Development (OECD). 1990. Labour Market Policies for the 1990s. Paris.

Psacharopoulos, George. 1991. From manpower planning to labour market analysis. International Labour Review 130(4).

Toffler, Alvin. 1970. Future Shock. New York: Random House. Wabe, J.S. 1974. Issues in manpower forecasting. In J.S. Wabe (ed.), Problems m Manpower

Forecasting. Westmead: Saxon House. Wallis, KJ. (ed.). 1987. Models of the UK Economy. A Fourth Review by the ESRC

Macroeconomic Modelling Bureau. New York: Oxford University Press.

205



labour market forecasting in australia: the science of the art

Documents