Forecasting,impact analysis and uncertainty propagation in regional integrated models: A case study of Australia

Abstract

The integration of input–output and econometric models at regional level has gained popularity for its superior performance in forecasting employment and examining the impacts of policies. There are a number of approaches to integrate the two models. This paper examines the integration of input–output with econometric modelling using two merging methodologies, namely coupling and holistic embedding. Each methodology is analysed with respect to the accuracy of its results of total and sectoral employment forecasting. Both methodologies are applied to a regional economy in Australia. The methodology which shows superior forecasting accuracy is applied to examine the significance of sectors that generate the highest number of employments relative to other sectors.

Keywords

Applied econometrics input–output analysis regional econometric modelling regional economics

Introduction

There are several methods for an input–output (IO) analysis¹ to be merged into an econometric model. In early attempts, IO models and econometric models were simply linked or conjoined (Kort and Cartwright, 1981; Kort et al., 1986; L’Esperance et al., 1977; Lienesch and Kort, 1992; Stevens et al., 1981; Stevens et al., 1983). However, as the integration framework has developed, innovative approaches have embedded IO coefficients in regional econometric models (Coomes et al., 1991; Moghadam and Ballard, 1988) or coupled the two models (Treyz et al., 1992). Some researchers also used IO coefficients as guidance in specifying equation of vector autoregression² (VAR) models (Fawson and Criddle, 1994). Moreover, a few researchers have incorporated IO linkages into Bayesian vector autoregression (BVAR) forecasting models (LeSage and Magura, 1991; Magura, 1990).

In a linked approach, the endogenous feedbacks between IO and econometrics are omitted due to the singular direction between the components of the two models. Alternatively, in an embedded³ approach, the presence of IO coefficients in econometric model is interchanged with a macro-regional variable that can remain the same across all sectoral equations (Rey, 1998). The a priori IO components in an econometric model are designed to improve the predictive accuracy of the embedded model. Most embedded approaches have been applied to model the economy of metropolitan areas (Coomes et al., 1991) or of regions consisting of one or more countries (Fawson and Criddle, 1994; Rey, 1998, 2000). More recently, Rickman (2002) has applied an integrating strategy to a state-level economy. Moreover, in a coupled approach, the final demand component of the IO analysis remains across the sectoral interactions and then is imposed on the econometric model (Treyz et al., 1992). As a result, the presence of errors in IO components in a coupled model will result in poor forecasting performance (Rey, 1998).

A large number of studies in the literature have focused on highlighting the merits of the integrated framework compared to the standalone traditional IO model and the standalone econometric model. On account of the comparative advantages of the integrated framework, over the standalone IO and econometric models, studies on the application of the integrated approach has proliferated over the past three decades. The literature review shows that with the exception of Rey (1998) that compares the alternative integration strategies for one region; all studies have applied a single integration approach for a certain region at a time rather than using a number of approaches to one region and comparing them. At the same time, considering the location specifics of different studies, literature shows that except for a few studies for the state of Queensland (West, 1991, 1995; West and Jackson, 1998), most of the studies have been performed for economically well-established regions within the US.

Having said that, the significant gap which appears in the literature is that we do not know how each integration approach works in a region which is in an economic transition with a shifting economic structure. The Illawarra region in New South Wales, Australia is a region which is facing a significant transformation of the economic structure. In this study, two methodologies are applied to integrate IO analysis with econometric framework in order to examine the Illawarra economy. The resulting two models are employed to forecast total and sectoral employment, and the superior model is then applied to analyse sectoral impacts on the regional economy. The results of the two models are further compared with results of a regional IO table of Illawarra and assessed on how well each model performs subject to cost and data availability. This paper leads the way for future studies which merge IO with econometric analysis with a view to building models that fit regions in economic transition. Using the two integration methodologies, IO components of final demand, export and sectoral linkages are incorporated into forecasting equations.

The region

Illawarra is a region in the Australian state of New South Wales (NSW). It is a coastal region situated south of Sydney and north of the Shoalhaven (a coastal region in southern NSW). The region encompasses an area of 1128 square kilometres and includes the three local government areas (LGAs) of Wollongong, Shellharbour and Kiama with an overall population of approximately 296,845 as of 2015 (ABS, 2016). The region experienced the second largest population growth in NSW, up by 3400 people during 2014–15 (ABS, 2016). The city of Wollongong is the third largest urban area in NSW (Figure 1).

Figure 1.

The Illawarra Map.

The main industries in the area were traditionally farming, followed by coal mining and lastly steel making. Nonetheless, the traditional manufacturing-based economic structure has been transforming into a knowledge orientation economic structure as described in the following context. Australia’s largest steel manufacturer, BlueScope Steel, operates at Port Kembla, located in south of the Illawarra. The high value of Australian dollar,⁴ the high costs of production due to lack of iron ore in the region and the international downturn in steel demand had a negative impact on the demand for steel from the BlueScope Steel plant. The reduced international demand and increasing automation of the steelmaking process led to closure of one of the two blast furnaces operating in Port Kembla plant (Binsted, 2015). The closure resulted in significant reductions in the labour needs at the plant and shed 1000 jobs (direct and indirect). The area, especially around Port Kembla and Wollongong, was once mainly known for mining and steel jobs, but since 2000s Retail Trade, Health Care & Social Services, Education & Training, and Professional, Scientific & Technical Services have played an increasingly important role in the region, overshadowing mining and steel manufacturing in terms of employment.

Wollongong is considered as an industrial hub in Australia (Gibson and Warren, 2013). Early economic activities focused on coal mining, rural processing and wood products. During mid-to-late 1800s, brickworks, dairying and meat processing, flour mills, salt works and saw mills activities were established. Although such businesses initially catered to local consumption, they gradually increased their supplies into the rest of NSW, including Sydney (Gibson and Warren, 2016). The producing sectors in Wollongong have traditionally benefited from the region’s proximity to Sydney’s much larger metropolitan economy (Horridge, 2011).

The Illawarra region has a significant youth unemployment rate and an overall unemployment rate of 6.7% as of December 2016 (LMIP, 2017), which is greater than the state average of 5.2%. The decline in steel industry has escalated diversification and development of new industries as a common need for the region. This has become particularly important for the Shellharbour LGA, which is overrepresented in unskilled and semi-skilled job categories. Job creation for unskilled and semi-skilled workers in the Shellharbour LGA has become critical to ensure the economic wellbeing of the region.

In terms of employment type in the region, nearly 21% of the FTEs are professionals, close to 16% are technicians and trade workers, 14% in clerical and administrative occupations, 11.5% are community and personal services workers, 10% managers, 9.5% sales, 9.2 labourers and 7.7% are machinery operators or drivers as of 2011 (ABS, 2012). The number of machinery operators and technicians, however, has declined by 1000 jobs⁵ as a result of 2011 closure of BlueScope blast furnace. As the unemployment levels increased as a result of the downscaling steelworks of BlueScope Steel, the regional economy was branded by high levels of unemployment, pollution and union’s constant battle for economic and political change (Gibson and Warrant, 2013).

Moreover, the expansion of the Port Kembla, one of country’s deep water ports and one of the only two ports in NSW, is a particular opportunity for the region. The port is Australia’s largest vehicle import hub, largest export grain terminal and the second largest coal export in NSW. There is also a partially constructed railway line, called Maldon-Dombarton, which will give the port direct access to south-western Sydney. Completion of this rail line will increase job opportunities for the Illawarra. The rail line also allows export freight from western NSW (primarily coal and grain) to access the port without depending on curfews or the need to use the South Coast rail line. This will reduce the number of freight movements on the South Coast line.

There has been a noticeable economic shift from heavy industry manufacturing to knowledge-oriented sectors and business service.⁶ This restructuring leads to perceptible incentives for applying sophisticated frameworks to model inter-sectoral interactions while capture temporal changes to analyse and plan the Illawarra economy. This framework is built around the Illawarra economy, albeit it is applicable to any regional or state economy where industrial restructuring is taking place and there is a need for robust planning. Accuracy in results of the analysis leads to more appropriate planning, which in turn helps sectoral diversification. The sectoral diversification leads to job creation, thus mitigating regional disparity between urban and regional areas. Regions similar to the Illawarra where unbalanced spatial structures are manifested in different conditions of life, unequal economic and development potentials can benefit from such frameworks. The results of the framework help identify sectors on which public investments can lead to enhanced economic development, job creation, thus increased employment. In accordance therefore, attention is now redirected to the objectives for merging IO with econometric analyses to investigate the Illawarra economy.

The objectives for merging IO analysis with econometric framework

Due to the wide variety of regional integrated models, it is vital to classify and specify integrated modelling within the context of this study. There are a number of integrated models in the literature. Five main integration categories are as follows:

Integrated models which apply entropy (probability) maximising methods, particularly in transport modelling (Wilson, 1984). Such models have also been generalized to analyse interregional commodity flows (Batten, 1983).

Integrated models which apply linear programming methods in interregional trade analysis (Harris, 1980). Such models have also been applied to regional environmental–economic interactions (Hafkamp and Nijkamp, 1981).

Models which apply generalization of IO such as the Batey and Madden (1986) frameworks and social accounting matrices (SAMs).

Computable general equilibrium (CGE) models which embrace price endogeneity as a way of extending the placement of IO relationships in a broader framework (Dixon and Jorgenson, 2013). Also, the enormous regional model (TERM), a “bottom-up” CGE model developed for Australia, offers a comprehensive integrated framework in the CGE category (Horridge, 2011).

Quantitative methodologies through which IO and econometric models are merged. Even within the integrated IO-econometric framework, there is a comparison analysis between the properties of the integrated IO-econometric modelling and its individual components, namely econometric and IO (Beaumont, 1990; Masouman and Harvie, 2018; Rey, 1998). One of the other recent examples in this category is the fully interregional dynamic econometric long-term IO (FIDELIO) system that has been developed for Europe, in which an IO module is embedded in a very extensive econometric framework (Kratena et al., 2013).

Before delving into delineation of the integrated IO-econometric models, it is vital to address the two-fold objective of integration for this study.

Theoretical objectives

The key theoretical objective for integrating IO with econometric modelling is to eliminate the limitations of each model. Although both IO analysis and econometric modelling are macroeconomic in core, IO analysis is static in nature whilst econometric modelling is dynamic.

The fundamental identity to begin the integration of the IO and econometric models with is the following identity

A^{ILW} . {GRO}^{ILW} + {FD}^{ILW} = {GRO}^{ILW}

(1)

where

GRO

is an n element vector of industrial gross regional output in the Illawarra;

FD

is an n element vector of aggregate industrial final demand expenditure in the Illawarra; and A is an nxn matrix of intermediate input requirements (coefficients, i.e.

a_{ij}

x_{ij} / X_{j}

) for the Illawarra. The equation (1) IO equation is then rearranged in the following format to calculate a change in gross regional output as a response to a change in final demand.

{GRO}^{ILW} = (I - A^{ILW})^{- 1} . {FD}^{ILW}

(2)

In the integration framework, the role of aggregation should be highlighted. There are m elements of aggregate final demand at a macro level, namely personal consumption C, investment I, government expenditures G, and net exports (Rey, 1998). Every one of the aggregate components is derived from the sum of the industry specific values.

Moreover, the restrictive assumption of productivity ratio and constant returns to scale are addressed in the integrated framework by defining a labour demand block in the econometric module. The correlation between labour demand and industry output is specified in econometric equations to track the time-path of labour demand shifts in the economy (Conway, 1990; West, 1991). Lastly, as stated earlier, IO is deterministic in nature with fixed coefficients that assume no substitutes for products, no uncertainty and infinite supplies. However, all such deterministic restrictions are well-defined econometrically with an added room for uncertainty. There are a number of studies included in the theoretical objectives that are presented in the supplemental file for this study.

Practical objectives

There are three practical objectives for integration. In view of the aforementioned economic transitions that have occurred in the Illawarra, there is a critical need for a framework that is capable of providing accurate impact analysis and robust forecasting in order to plan ahead. Therefore, three crucial desiderata that we look forward to achieving by merging the two models are improved impact analysis, more accurate forecasting, and less measurement errors (Rey, 1998).

In terms of impact analysis, as mentioned in the previous section, given the static nature of IO analysis, its main limitation is that it is constructed based on data from a single year inter-sectoral relationships, thus not capturing changes through time. This limitation is addressed by coupling the inter-sectoral relationships with the dynamics of econometric modelling. Contrariwise, the detailed inter-sectoral representation of IO is missing in a standalone regional econometric model. This lack is equally addressed by embedding the IO’s detailed a priori data into econometric equations. Henceforward, a dynamic and industrially disaggregated tool can be developed (Moghadam and Ballard, 1988; Motii, 2005; Rey, 1994; West, 1991).

In terms of forecasting accuracy, the integrated framework has been suggested by a number of modellers (Glennon et al., 1987; Kratena et al., 2013; Moghadam and Ballard 1988; Motii, 2005; Rey, 1998; West, 1991, 1995). In this regard, coupling the time-series estimates with the inter-sectoral coefficients and embedding IO coefficients within econometric equations can address the deterministic restriction of IO by generating confidence intervals and point estimates. Similarly, interdependencies among economic factors such as unemployment, labour force participation rate, etc., in labour market are analysed through a VAR approach. The generated estimates are then assigned prior probabilities and treated stochastically by a Bayesian VAR to determine sectoral employments (Fawson and Criddle, 1994; LeSage and Magura, 1991; Magura, 1987; Motii, 2005; Rey, 1998).

Lastly, high costs⁷ associated with conducting surveys together with the usual low response rate of surveys, constrain update on and availability of data used in IO tables, particularly at a subnational level⁸ (Lahr and Stevens, 2002; West, 1991). The issue of data obsoleteness is commonly addressed by regionalizing the national coefficients to estimate regional coefficients, known as regionalization method (Rey, 1998).

In the light of the theoretical and practical objectives, and considering the evolving economic transformations in the Illawarra, two of the popular methodologies for merging IO with econometric modelling are extended and applied in this study to investigate the Illawarra economic transitions. The compilation procedure for application of the methodologies is presented in the following section.

Compilation of merging methodologies to investigate the Illawarra economy

Having developed a standalone regional IO model⁹ for the Illawarra economy, the focus of this paper turns to integration of IO model with an econometric framework using two methodologies, namely coupling and embedding. As mentioned previously, there has been a plethora of studies on the integrated framework in recent years. A number of methodological issues, associated with the merits of the intergraded framework versus traditional models, have been identified in the literature. As per the main discussion of this paper, there are a number of methods for integrating the two models, yet with the exception of Rey (1998), there is no evidence on the comparative performance and characteristics of the different methods.

There have been some constraints equated with modelling at the regional scale. Although there have been a number of cases which have applied these models and provided useful insights (Beaumont, 1990; Kort and Cartwright, 1981; Kort et al., 1986), they have mainly applied the integration on the national or state level modelling, where disaggregated cross-section and time-series data are archetypally available. Conversely, there has been no attention given to the comparative properties of different integrated models in practice (Rey, 1998). To fill in this gap, we merge estimates of an econometric set of equations into the Illawarra IO model (coupled methodology). Then, we embed an IO module is into a host econometric framework (embedded methodology). Both methodologies are applied to the Illawarra regional economy and the results of both are compared in order to shed some lights on the relative properties of the two methodologies. Finally, we investigate the Illawarra economy by analysing the impact of $1 million expenditure on three prominent sectors, using a coupled model.

Coupling

The highest use of time-series data is attributed to the coupled methodology. The econometric component of this methodology is composed of the following three main blocks:

Consumption

Investment (government/private expenditure)

Income

Each block contains a set of regional econometric equations and the results of each block are incorporated into the sectoral employment, sectoral output and final demand vector of the Illawarra IO model. The consumption block is the first part of integration and specifies consumption expenditure as a function of regional disposable income, national interest rate and unemployment trends (West, 1991). There are five elements in the final demand vector. Three of the five final demand elements are endogenized in the econometric equations in the coupled model. These elements are: (1) private final consumption expenditure; (2) government final consumption expenditure and (3) gross fixed capital formation. The remaining two elements, namely, exports and imports are exogenous variables.

Borrowing the underlying assumptions partly from the regional economic models, Inc. (REMI) by Treyz et al. (1991) and partly from Washington Projection and Simulation Model (WPSM) by Conway (1990) the sectoral deliveries are directed to three of the five elements of final demand. Although REMI (Treyz et al., 1991) is a model of Walrasian origin and WPSM (Conway, 1990) is one of Marshallian, their treatments of sectoral deliveries to final demand vector are complementary. The aggregate component of the final demand is first implemented in an equation as a random variable and is estimated as a function of related determinants. Using a matrix of regional final demand distribution, these aggregate values are then dispersed to individual sectors (Rey, 1998).

Total regional private consumption expenditures (PCE) are modelled by applying similar underlying assumptions from REMI and WPSM (Conway, 1990; Treyz et al., 1991). The determinants of PCE are regional disposable income RDY, the national interest rate NIR and the regional unemployment rate RUR

{PCE}_{t}^{ILW} = f_{PCE} ({RDY}_{t}^{ILW}, {NIR}_{t}^{ILW}, {RUR}_{t}^{ILW})

(3)

The social security transfer payments are separated out of RDY based on the national proportions. This is done so to allow for differences in the marginal propensities to consume (MPC) out of the two income streams. The NIR is intended to capture the effects of financing costs on durable goods purchases, while the RUR is used as a proxy for consumer confidence.

To begin with, we specify regional investment as a function of regional capital stocks. Net investment ${NI}_{t}$ is specified by obtaining the difference between the current capital stock and the lagged levels of the capital stock

{NI}_{t}^{ILW} = λ ({RKS}_{t}^{ILW} - {RKS}_{t - 1}^{ILW})

(4)

where λ measures how fast the lagged capital stock levels have changed to the current capital stock levels. We specify the current level of capital stock applying a fixed capital to output ratio, denoted by µ

{RKS}_{t}^{ILW} = μ ({GRP}_{t}^{ILW})

(5)

And substituting equation (4) into equation (5) yields

{NI}_{t}^{ILW} = λ μ {GRP}_{t}^{ILW} - λ {RKS}_{t - 1}^{ILW}

(6)

We use equation (6) and extend it by adding more elements in order to model gross investment, business inventories, government expenditure, personal income, disposable income and transfer payments. All the equations for coupling model are presented in the supplemental file for this study.

Generic embedding

Of the two methodologies in the integration framework, the embedded methodology uses the least level of time-series data, thus it is easier and less complex than the coupled methodology to build. This methodology highlights the inter-industry relations which capture the employment demand variables of the IO model within an econometric equation.

The embedded model in this study is a generic extension from Moghadam and Ballad’s (1988), Rey’s (1998) and Motii’s (2005) work. The main contribution of this part of the study to the literature is twofold. Firstly, the practical novelty in that this is the first embedded econometric IO model designed for and applied to the Illawarra economy and in a broader context, to the regional Australia.¹⁰ Secondly, this is the first study which compares the embedded methodology with the coupled methodology to the same region in Australia. The overall algebraic notation is based on the following time-series specification

{EMP}_{t} = β_{0} + β_{1} V_{t} + β_{2} Z_{t} + β_{3} {ID}_{t} + ɛ_{t}

(7)

where

{EMP}_{t}

is employment at time t, estimated by a function of local V_t, external Z_t and inter-industry

{ID}_{t}

variables. V_t represents local macro economy variables, both endogenous and exogenous, such as total personal income, population, wage rates, etc.Z_t represents national and other external variables that can impact the region. These variables can be either sectoral estimated coefficients that establish the elasticity between the region and the nation as well as certain policy variables that are important in industries where the Illawarra significantly differs from the rest of the Australia, such as steel or the education sectors. Finally,

{ID}_{t}

represents inter-sectoral values and IO linkages, which measures the demand for the output of one industry from the other industries within the region.

Data sources

The choice of sectors used in the econometric model is determined by the availability of sectoral time-series data on Australian Bureau of Statistics (ABS) and Labour Market Information Portal (LMIP). The choice of sectors for the IO table is determined by the availability of a consistent set of cross-sectional data for a number of variables at the sectoral level, including gross regional products, wages and salaries and employment. The time-series data used for the forecasting experiments and the econometric equations are for the 2001–2011 period.

The primary data sources include the State Accounts, New South Wales Yearbook, Labour Force Statistics, Manufacturing Statistics, Consumer Price Index, IRIS Annual Publications, plus other miscellaneous publications such as Census. The initial IO table was constructed in the School of accounting, economics and finance at the University of Wollongong. The initial table is updated using a hybrid methodology, which is a combination of survey and estimated data.

Finally, it is important to note that based on the nature of data that is available for this region, the results of each methodology can be gauged and evaluated to fit the nature and availability of data for other regions in future studies.

Comparison of the estimation results

For the first experiment, the two integration methodologies, namely the coupled and the generic embedding, are applied to a dynamic ex-post scenario to forecast sectoral employments for the period of 2010–11.¹¹ Each of the two methodologies in this experiment is based on parameters estimated over the 2001–2011 sample period. The comparison is made on the basis of percentage error (PE) for the predicted values of each variable over the last one year of the sample period.

The results of the forecast show that the embedding methodology performs better in seven sectors than the coupled methodology with respect to forecasting sectoral employment. This is due to the even distribution of the estimated values across all sectors, which is an advantage achieved by embedding the IDV within an econometric framework, provided that there is sectoral time-series available. As noted on Table 1, the PE of the embedded methodology is far less than that of the coupled on the following sectors:

Agriculture, Forestry & Fishing

Wholesale Trade

Retail Trade

Accommodation & Food Services

Professional, Scientific & Technical Services

Public Administration & Safety

Arts & Recreation Services

Table 1.

Sectoral forecast results (2011 Sectoral employment: actual vs. integrated methodologies).

No.	Industry of employment	Actual	Coupled	PE	Embedded	PE
1	Agriculture, forestry and fishing	498	573.4	0.152	455.1	−0.086
2	Mining	3057	2806.1	−0.082	2801.0	−0.084
3	Manufacturing	11,938	12,728.9	0.066	10,982.0	−0.080
4	Electricity, gas, water and waste services	1326	1226.2	−0.075	1221.3	−0.079
5	Construction	9392	8802.5	−0.063	10,978.3	0.169
6	Wholesale trade	3099	3337.2	0.077	3285.8	0.060
7	Retail trade	12,698	12,337.6	−0.028	12,906.3	0.016
8	Accommodation and food services	8733	9394.6	0.076	9224.8	0.056
9	Transport, postal and warehousing	6100	7028.1	0.152	7072.6	0.159
10	Information media and telecommunications	1474	1428.3	−0.031	1562.4	0.060
11	Financial and insurance services	4303	4334.3	0.007	4809.8	0.118
12	Rental, hiring and real-estate services	1858	1961.1	0.056	2094.4	0.127
13	Professional, scientific and technical services	6229	5717.3	−0.082	6343.4	0.018
14	Administrative and support services	3760	3491.6	−0.071	3240.0	−0.138
15	Public administration and safety	8445	8959.1	0.061	8112.9	−0.039
16	Education and training	12,482	13,355.4	0.070	10,372.6	−0.169
17	Healthcare and social assistance	16,026	15,306.6	−0.045	13,513.4	−0.157
18	Arts and recreation services	1704	1960.3	0.151	1636.9	−0.039
19	Other services	4883	5196.5	0.064	3874.9	−0.206
	Total	118,004	119,945.04		114,487.98

Source: Computed by the author.

PE: percentage error.

One inference that can be surmised from the type of sectors that exhibit the least error using embedded methodology is that they are the sectors often noted for hiring part-time employees and at the same time, showing the highest interactions with consumers and the household sector. Therefore, since the a priori information from the IO is incorporated in each industry time-series data, the result can be more accurate. This is also the reason behind the underestimation of the total employment forecast yielded by the embedded methodology.

The coupled methodology, however, outperforms the embedded on all the other 12 sectors as well as the total employment forecast. The average PE for the coupled methodology is 0.024 with a standard deviation of 0.0823 while the average PE for the embedding model is −0.015 with a standard deviation of 0.1144. A likely justification for this is that in a more diversified economy it is critical to avail of inter-sectoral linkages in defining the employment demand equations. The coupled methodology imposes an exhaustive set of inter-sectoral relations in each employment equation and an extensive series of final demand equations; it is more sector specific in terms of forecasting. To provide a more visual depiction of the comparison, the sectoral PE’s are presented in graphs in Figure 2.

Figure 2.

Comparison of the percentage error for coupled and embedded forecasts.

One may argue that selecting one specific year is not a suitable metric for comparison. There is the possibility of economic shocks and major structural changes taking place in such a short period, hence casting a shadow over the comparative performance of a given model. As a result, we decided to compare the results by back testing the total employment forecast over the sample period 2001–2011 for each model and compared against the actual figures.

As noted in Table 2, once again the embedded model exhibits underestimation of the total employment forecast for 7 out of 11 observations with an overall standard deviation of 0.085. On the other hand, the coupled model provides more accurate forecasts of the total employment for the 11 observations by a standard deviation of 0.049 and underestimation for 5 out of 11 observations. Once again, in order to provide a more visual depiction, the results of the back testing are presented in form of a graph in Figure 3.

Table 2.

Total employment back testing forecasts, 2001–2011.

Year	Actual employment	Coupled	PE	Embedded	PE
2001	103,826	89,919.87	−13.39%	110,748.02	6.67%
2002	104,962	104,911.88	−0.05%	96,114.88	−8.43%
2003	106,099	106,407.50	0.29%	101,033.22	−4.77%
2004	107,235	108,909.90	1.56%	102,359.24	−4.55%
2005	108,372	99,169.05	−8.49%	113,866.16	5.07%
2006	109,508	111,010.09	1.37%	98,171.17	−10.35%
2007	111,207	104,854.00	−5.71%	127,887.68	15.00%
2008	112,906	112,021.96	−0.78%	121,136.93	7.29%
2009	114,606	114,730.81	0.11%	106,564.10	−7.02%
2010	116,305	116,461.40	0.13%	103,743.43	−10.80%
2011	118,004	119,945.04	1.64%	114,487.98	−2.98%

Source: Computed by the author.

PE: percentage error.

Figure 3.

Graphs for total employment back testing.

The second experiment is an applied impact analysis on the Illawarra economy. For this experiment, we only use a coupled methodology, attributable to its superior performance from the earlier experiment. In this experiment, the objective is to estimate impact of a hypothetical AU$1 million increase in final demand on three sectors, namely (a) Health Care & Social Services with 15,708 employments; (b) Retail Trade with 12,446 employments; and (c) Education & Training with 12,234 employments. The aforementioned sectors are selected on the grounds that they had the highest number of FTEs by the end of sample period (2011). Tables 3 to 5 show the results of these impact analyses arisen from change in the final demand based on the coupled IO-econometric model. There are only seven sectors displayed on each table out of the 30 sectors designed for the IO sectoral profile. The seven sectors are selected on the basis of the significance of the total effect on them relative to the other 23 sectors, i.e. the sectors with the highest total effects are selected and displayed on each table. Description of all the characters that are used Tables 3 to 5 are listed in the supplemental material.

Table 3.

Seven highest impacted sectors as final demand increases in Health Care.

No.	Sector	Final demand	Industrial supply	Consumption	Total	Percent 1	Flow-on	Percent 2
1	Health care and social assistance services	1	0	0	1	46.6	0	3.3
2	Finance and insurance services	0	0	0.2	0.2	10.5	0.2	19
3	Accommodation and food services	0	0	0.1	0.1	5	0.1	9.1
4	Transport, postal and warehousing	0	0	0	0.1	2.3	0.1	4.2
5	Ownership of dwellings	0	0	0.1	0.1	6.1	0.1	11.1
6	Professional, scientific and technical services	0	0	0	0.1	2.4	0.1	4.4
7	Education and training	0	0	0.1	0.1	2.3	0.1	4.2
	Total	1	0.2	1	2.2	100	1.2	100
	Multiplier	1			2.234		1.234

Source: Computed by the author.

Note: All dollar figures are in millions.

The first set of findings pertains to the AU$1 million increase in Health Care & Social Services as the largest employing sector in the region (ABS, 2011). In terms of indirect effect, the results show that the total immediate inter-industrial impact would be $200,000. In terms of induced effect, there would be $200,000 increase in final demand for Finance & Insurance sector, and $100,000 increase in final demand for each of the (a) Accommodation & Food Services, (b) Ownership of Dwellings, and (c) Professional, Scientific & Technical Services sectors. With respect to Type II Multiplier proportions (direct + indirect + induced effects), Health Care & Social Services is affected by 46.6%, Finance & Insurance Services by 10.5%, and the third highest being affected is the Ownership of Dwellings which would be affected by 6.1%. The type II multiplier is 2.234, which means the total impact of a AU$1 million increase on Health Care & Social Services would yield $AU2.234 million increase (including the AU$1 million) across all the sectors within the Illawarra economy. The aforementioned findings are presented in Table 3.

The second set of findings relates to the final demand increase in Retail Trade sector, the second largest employer in the Illawarra. In terms of the indirect impacts, the inter-sectoral output for Finance & Insurance Services, and Professional, Scientific & Technical Services would increase by $100,000 each. In terms of the induced effect, there would be $100,000 increase in the final demand for each of the Retail Trade, Finance & Insurance Services, Accommodation & Food Services, and Ownership of Dwellings sectors. With respect to the proportion of the total effects (Percent 1), the top three sectors would be Retail Trade with 50.2%, followed by Finance & Insurance Services with 9.3%, and lastly, Ownership of Dwellings with 4.6% of the total effect. To sum it all, the Type II Multiplier is 2.211, which means that the AU$1 million increase in the final demand of the Retail Trade would entail another AU$1.211 million increase in direct and indirect effects across all the sectors in the Illawarra economy. The aforementioned results are presented in Table 4.

Table 4.

Seven highest impacted sectors as final demand increases in Retail Trade.

No.	Sector	Final demand	Industrial supply	Consumption	Total	Percent 1	Flow-on	Percent 2
1	Retail trade	1	0	0.1	1.1	50.2	0.1	9.1
2	Finance and insurance services	0	0.1	0.1	0.2	9.3	0.2	17
3	Accommodation and food services	0	0	0.1	0.1	4.2	0.1	7.6
4	Transport, postal and warehousing	0	0	0	0.1	2.5	0.1	4.5
5	Rental, hiring and real estate services	0	0	0	0.1	2.7	0.1	5
6	Ownership of dwellings	0	0	0.1	0.1	4.6	0.1	8.4
7	Professional, scientific and technical services	0	0.1	0	0.1	4.5	0.1	8.2
	Total	1	0.4	0.8	2.2	100	1.2	100
	Multiplier	1			2.211		1.211

Source: Computed by the author.

Note: All dollar figures are in thousands.

Table 5.

Seven highest impacted sectors as final demand increases in Education.

No.	Sector	Final demand	Industrial supply	Consumption	Total	Percent 1	Flow-on	Percent 2
1	Education and training	1	0	0	1	56.2	0	4.3
2	Finance and insurance services	0	0	0.1	0.2	8.1	0.2	17.8
3	Retail trade	0	0	0.1	0.1	4.7	0.1	10.2
4	Accommodation and food services	0	0	0.1	0.1	4	0.1	8.6
5	Ownership of dwellings	0	0	0.1	0.1	4.4	0.1	9.7
6	Repairs	0	0	0	0	1.1	0	2.4
7	Personal and other services	0	0	0	0	0.7	0	1.6
	Total	1	0.2	0.6	1.8	100	0.8	100
	Multiplier	1			1.845		0.845

Source: Computed by the author.

Note: All dollar figures are in thousands.

The final set of findings pertains to the final demand increase in Education & Training by AU$1 million. As exhibited in Table 5, the sum of the indirect impacts for this set of findings is $200,000. In term of induced effects, the consumption for the products and services of the three sectors, namely Finance & Insurance Services, Accommodation & Food Services, and Retail Trade would increase by $100,000 each and in total, there would be $600,000 increase in induced effects in the economy. In terms of the proportions of the total effect, Education & Training would be affected by 56.2%, followed by Finance & Insurance Services by 8.1%, and lastly, Retail Trade by 4.7%. The Type II multiplier indicates a further $845,000 indirect and induced increase across different sectors within the economy as a result of AU$1 million increase in the final demand of the Education & Training sector. Table 5 presents the aforementioned results.

Overall, among the three top employing sectors, the sector which yields the largest Type II Multiplier as a result of the AU$1 million surge is Health Care & Social Services. A million dollar increase of government/private expenditure on this sector increases another $1.234 million in total increase across other sectors within the economy.

A note on estimation results

At the outset, it is noteworthy that a baseline comparison of the two integrated models, the standalone IO and regional econometric models have not been presented in this paper. Although such a comparison helps assess the difference in the extent of further improvement in prediction accuracy generated by the two integrated models, it is beyond the scope of this paper. However, a baseline comparison of a coupled model and a standalone IO is conducted by Masouman and Harvie (2018) and in a different study several versions of an embedded model are compared with a standalone IO analysis (Masouman and Harvie, 2017).

With regard to the econometric estimates, if the sample size is relatively small, as is the current case, it is recommended to use a two-step method to estimate generalised least squares (Gujarati and Porter, 2009). Step 1 involves obtaining an estimate of the unknown ρ and step 2 involves using that estimate to transform the variables to estimate the generalized difference equation, or the generalized least squares (GLS). However, instead of using true ρ, we use $\hat{ρ}$ in this study and such a method of estimation is known in the literature as feasible GLS (FGLS) or estimated GLS (EGLS) methods. The small sample size can lead to correlation between the residuals in the regression and ordinary least square in such a case can be statistically inefficient.

Moreover, because the sample size has impacts on estimated parameters in terms of precision and accuracy, a setback is that the sample size effects might be different for the two econometric models used in the embedded and coupled approaches. Therefore, we use a Monte Carlo simulation to run 50,000 iterations for each set of the macro equations in the coupled model and each industry in the embedded model. The results of the simulation are presented in Tables 6 and 7 and a discussion of the results of the simulations is provided, both of which are in the supplemental material.

Conclusion

This paper has investigated the comparative properties of two integration methodologies for combining IO analysis with econometric modelling. According to the studies found in literature, integrated models are noted to be superior in terms of forecasting and impact analysis compared to traditional IO or econometric models (Conway, 1990; Motii, 2005; Rey, 1998; West, 1991). Nevertheless, findings of this study suggest that there is also a wide variation in the manner each methodology is developed, which is also evident from an earlier study of this kind (Rey, 1998).

A plethora of integration models for regional analysis tools have been developed since the inception of regional science as a field. However, except for Rey (1998), a comparison of the fundamental properties of the different methodologies to integrate IO with econometric has not been addressed. As accurately postulated by Rey (1998), integrating IO analysis and econometric modelling requires two critical factors to be pondered thoroughly: (a) integration regime and (b) integration structure.

Regarding the properties of each of the integrated methodologies, coupled model dominates the comparison either when the objective is forecasting sectoral employment or total employment. As noted earlier, this methodology is highly data intensive and, as the findings indicate, there is no direct relationship between the complexity of this approach and its performance. Hence, the trade-off between the cost¹² of developing the model and its performance appears to be ambiguous. In an overly diversified economy where inter-industry structure is highly developed and endogenous factors are central in structuring the economy – and importantly, data availability is not an issue – a coupled model seems to be a suitable choice for forecasting employment. With respect to impact analysis, the impacts are fully spread across industries within regional economy. The estimated impacts also tend to vary less across different sectors.

Lastly, it is also worthwhile to note that after running the econometric module for the coupling model, we used the Leontief inverse in order to extend the impacts of final demand change across other disaggregated sectors in the IO analysis. In general, a coupling methodology helps specify sectoral employment as a function of sectoral output and labour productivity. In contrast, an embedding methodology helps specify employment in a fashion similar to regular econometric modelling. Accordingly, the estimate variance in the embedded model is more commensurate with the econometric modelling while this variance in the coupling model is in accordance with the detailed sectoral disaggregation of the IO analysis while retaining the dynamics of the econometric model to a higher extent than the embedded model.

Supplemental Material

Supplemental material for Forecasting, impact analysis and uncertainty propagation in regional integrated models: A case study of Australia

Supplemental material for Forecasting, impact analysis and uncertainty propagation in regional integrated models: A case study of Australia by Ashkan Masouman and Charles Harvie in Environment and Planning B: Urban Analytics and City Science

Footnotes

Declaration of conflicting interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: This research has been conducted solely by the authors and the article is written independent of KPMG. The article is not related to any services provided by KPMG and does not represent modeling or research capabilities of KPMG.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Ashkan Masouman is an experienced management consultant with a demonstrated history of working in economics and investment related businesses. His professional specialisation is in financial modelling, corporate finance, policy analysis, microeconomics, and portfolio management. He is currently a consulting professional with a focus on econometrics and quantitative finance.

Charles Harvie has been involved in a number of research activities and projects for major international organisations including the Asian Productivity Organisation, Asian Development Bank, Asian Development Bank Institute, the Economic Research Institute for ASEAN and East Asia and the Department of Foreign Affairs and Trade, Australia.

References

Australian Bureau of Statistics (ABS) (2011) National Regional Profile, Wollongong, Catalogue 1379.0.55.001.

Australian Bureau of Statistics (ABS) (2012) Regional Population Growth, Australia, Catalogue 3218.0.

Australian Bureau of Statistics (ABS) (2016) Regional Statistics by ASGS, 2011-2016 Illawarra 107, Cat. no. 1379.0.55.001, ABS, Canberra.

Australian Bureau of Statistics (ABS) (2017) National Regional Profile, ABS, Canberra.

Batey

PWJ

Madden

(1986) Integrated Analysis of Regional Systems, London: Pion.

Batten D (1983) The role of entropy and information theory in spatial input-output modeling. Spatial Analysis of Interacting Economies. Netherlands: Springer.

Beaumont PM (1990) Supply and demand interaction in integrated econometric and input-output models. International Regional Science Review 13(2): 167–182.

Binsted T (2015) Union given eight weeks to save Port Kembla. The Sydney Morning Herald, 24 August 2015.

Conway

RSJR

(1990) The Washington projection and simulation model: A regional inter-industry econometric model. International Regional Science Review 13(1–2): 141–165.

10.

Coomes

Olson

Merchant

(1991) Using A metropolitan-area econometric model to analyse economic development proposals. Urban Studies 28: 369–382.

11.

Dixon PB and Jorgenson DW (2013) Handbook of computable general equilibrium modelling. North Holland 1: 1–1841.

12.

Fawson C and Criddle KR (1994) A comparative analysis of time series approaches to modelling intersectoral and intercounty employment linkages in rural regional labour markets. Journal of Regional Science 34: 57–74.

13.

Gibson C and Warren A (2016) Resource-Sensitive Global Production Networks: Reconfigured geographies of timber and acoustic guitar manufacturing. Economic Geography 92(4): 430–454.

14.

Glennon D, Lane J and Johnson S (1987) Regional econometric models that reflect labour market relations. International Journal of Forecasting 3(2): 299–312.

15.

Gujarati DN and Porter DC (2009) Basic econometrics (5th ed.). Boston: McGraw-Hill.

16.

Hafkamp W and Nijkamp P (1981) Multiobjective modelling for economic–environmental policies. Environment and Planning A: Economy and Space 13(1): 7–18.

17.

Harris CC (1980) Symposium on multiregional forecasting and policy models. Journal of Regional Science 20(2): 159–171.

18.

Horridge M (2011) The TERM model and its data base. The Centre of Policy Studies, General Paper No. G-219 July 2011.

19.

Kim

Hewings

GJD

(2009) An application of an integrated transport network – Multiregional CGE model to the calibration of synergy effects of highway investments. Economic Systems Research 21(4): 377–397.

20.

Kort

Cartwright

(1981) Modelling the multiregional economy: Integrating econometric and input-output models. The Review of Regional Studies 11: 1–17.

21.

Kort

Cartwright

Beemiller

(1986) Linking regional economic models for policy analysis. In: Perryman

Schmidt

(eds) Regional Econometric Modelling, Boston: Kluwer.

22.

Kratena K, Streicher G, Temurshoev U, et al. (2013) Fully Interregional Dynamic Econometric Long-term Input-output Model for the EU27. European Union: Joint Research Centre Scientific & Policy Reports, European Commission, Institute for Prospective Technological Studies, 2013.

23.

L’EsperanceWL King

Sines

(1977) Conjoining an Ohio input-output model with an econometric model of Ohio. Regional Science Perspectives 7: 54–77.

24.

Labour Market Information Portal (2017) Illawarra, Employment by Industry, February 2017. Available at: http://lmip.gov.au/.

25.

Lahr ML and Stevens BH (2002) A study of the role of regionalization in the generation of aggregation error in regional inputoutput models. Journal of Regional Science 42(3): 477–507.

26.

LeSage JP and Magura M (1991) Using Inter-industry Input-Output Relations as a Bayesian Prior in Employment Forecasting Models. International Journal of Forecasting 7(2): 231–238.

27.

Lienesch

Kort

(1992) The NRIES II multiregional macroeconomic model of the United States. International Regional Science Review 14: 255–274.

28.

Magura M (1987) The use of Input-output fables in specifying inter-industry and interregional labour market linkages. Papers in Regional Science 62(1): 117–123.

29.

Magura

(1990) Using input-output data in a Bayesian autoregressive forecasting model. In: Anselin

Madden

(eds) New Directions in Regional Analysis: Multi-Regional Approaches, London: Bellhaven, pp. 133–145.

30.

Masouman A and Harvie C (2017) Measuring economic change in the illawarra, new south wales: An integrated framework. Australian Economic Review 50: 294–308.

31.

Masouman A and Harvie C (2018) Regional economic modelling through an embedded econometricinter-industry framework. Regional Studies. 52(9): 1237–1249.

32.

Moghadam

Ballard

(1988) Small area modelling of the industrial sector (SAMIS): An integrated econometric-inter-industry approach. Environment And Planning A 20(5): 655–668.

33.

Motii

(2005) A dynamic integration approach in regional input-output and econometric models. The Review of Regional Studies 35(2): 139–160.

34.

Rey

(1998) The performance of alternative integration strategies for combining regional econometric and input-output models. International Regional Science Review 21(1): 1–36.

35.

Rey

(2000) Integrated regional econometric + input-output modelling: Issues and opportunities. Papers in Regional Science 79(3): 271–292.

36.

Rickman DS (2002) A bayesian forecasting approach to constructing regional input–output based employment multipliers*. Papers in Regional Science 81(4): 483–498.

37.

Stevens BH, Treyz GI and Kindahl JK (1981) Conjoining An Input-Output Model And A Policy Analysis Model: A Case Study Of The Regional Economic Effects Of Expanding A Port Facility. Environment and Planning A 13: 1029–1038.

38.

Stevens BH, Treyz GI, Ehrlich DJ, et al. (1983) A new technique for the construction of non-survey regional input-output models. International Regional Science Review 8: 271–286.

39.

Treyz G, Rickman DS and Shao G (1991) The REMI economic-demographic forecasting and simulation model. International Regional Science Review 14(3): 221–253.

40.

Treyz GI, Rickman DS and Shao G (1992) The REMI economic-demographic forecasting and simulation model. International Regional Science Review 14: 221–253.

41.

Warren A and Gibson C (2013) Making things in a high dollar Australia. Journal of Australian Political Economy 71(1): 26–50.

42.

West

(1982) Sensitivity and key sector analysis in input-output models. Australian Economic Papers 21(39): 365–378.

43.

West

(1991) A Queensland input-output econometric model: An overview. Australian Economic Papers 30(57): 221–240.

44.

West

(1995) Comparison of input-output, input-output + econometric and computable general equilibrium impact Models at the regional level. Economic Systems Research 7(2): 209–227.

45.

West

Jackson

(1998) Input-output + econometric and econometric + input-output: Model differences or different models? Journal of Regional Analysis and Policy 28(1): 33–48.

46.

Wilson AG (1984) Making urban models more realistic: some strategies for future research. Environment and Planning A 16(4): 1419–1432.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.83 MB