Contents (continued )
2. Expenditure Switching and Impact of the Introduction of EGMs
2.2 Time Series Econometric Data Analysis
2.2.2 EGM Introduction and Compositional Changes
Turning to the second question, even with stable consumption expenditure patterns overall, it is possible that this disguises a significant change in the composition of expenditure driven by EGM expenditure that may have impacted on businesses in specific sectors. In general
63 Of course this model is only looking at aggregate behaviour, so this does not mean that the consumption behaviour of some
economists treat any shift between sectors as irrelevant in terms of net economic impact. From the point of view of the welfare of the population as a whole, it doesn’t matter whether expenditure occurs in one sector (or on one good/service) rather than another, provided that this spread of expenditures represents the rational choices of the consumer rather than a response to compulsion (whether legal, illegal, or through addiction).
An activity can only have a net economic impact if it leads to a higher level of expenditure than would otherwise have been the case (either through attracting export income64, or through increasing the share of income consumers choose to spend,65 or if there is some externality that increases the efficiency of the economy). In this case, as a significant share of EGM expenditure comes from problem gamblers (the Productivity Commission estimated 41 per cent) it cannot necessarily be said to represent a rational choice, and therefore there are two potential sources of externality:
• the induced expenditure; and
• the social harms caused by problem gambling.
Hence, it is legitimate to seek to discover whether this ‘involuntary’ expenditure has come from expenditure switching from other sectors, reduced growth in expenditures in other sectors, or from reduced savings.
There are always difficulties in trying to get to grips with the ‘economic impact’ whether
in terms of employment or GDP of a policy change, as there are rarely good natural
experiments. We are never faced with a situation where the only thing changing is the policy choice. Instead, in most cases, almost everything is changing and it is necessary to use statistical techniques to try and establish whether there appears to be any additional impact of the policy.
The source for our detailed information on expenditure patterns is the components of state final demand in the national accounts (ABS 5206, Table 87), which split household consumption expenditure into a number of categories. In order to identify the impact of gambling we created two additional variables – “EGM gambling” and “other gambling” – from the Australian Gambling Statistics data (Office of Economic and Statistical Research, Queensland Treasury, 2005). To remove double counting these gambling expenditures were extracted from the series ‘recreation and culture’ where they are recorded. The categories of expenditure are set out below:
• Food;
• Cigarettes and tobacco;
• Alcoholic beverages;
• Clothing and footwear;
• Rent and other dwelling services;
• Electricity, gas and other fuel;
64 As one economist puts it in private correspondence to the researchers, “The message is that in the absence of export income, the
gambling industry simply acts to transfer income from everyday necessities to publicans and casino owners. In other words if you remove the gambling industry children get fed or adequate health care instead of pub owners buying luxury goods”.
65 This is likely to have an ambiguous effect, increasing short term growth but potentially leading to lower long-term growth as
lower savings either decreases investment directly, or in an open economy tends to increase interest rates as investment capital needs to be attracted from abroad.
• Furnishings and household equipment; • Health; • Purchase of vehicles; • Operation of vehicles; • Transport services; • Communications;
• Recreation and culture (minus net gambling expenditure);
• Education services;
• Hotels, cafes and restaurants;
• Insurance and other financial services;
• Other goods and services;
• EGM gambling; and
• Other gambling.
Graphing the rates of growth in categories of retail sales expenditure allows us to form initial views as to the existence or otherwise of any impact from the introduction of EGMs. As there is such a substantial list of expenditure categories they have been split into four groups for the purposes of this graphing (see Figures 2.2 to 2.5).
It is clear that expenditure between categories of consumption shifts strongly year to year (as would be expected given their relative prices are shifting). Although visual scans are by no means definitive, there does appear to have been a shift in ‘other gambling’ which occurred at the time of the introduction of EGMs and a shift in expenditure on hotels, cafes and restaurants (refer Figure 2.2).
Expenditure on ‘recreation and culture’ (minus net gambling expenditure) and ‘cigarettes and tobacco’ may have fallen after EGMs were introduced (see Figure 2.3), although this is uncertain. Whilst the changes are much less significant than that of ‘other gambling’, a visual scan of the charts also suggest the possibility that expenditure on ‘furnishings and household equipment’ (see Figure 2.4) may have fallen after the introduction of EGMs.
On a visual scan there does not appear to have been any shift in the other categories of consumption shown following the introduction of EGMs.
In order to test whether the intuition of our visual scans is correct we have applied statistical tests by running regressions of the expenditure on various categories of consumption to test whether the level of EGM expenditure was negatively correlated with expenditure. Of course as correlation does not prove causation, even a strong negative correlation between some category of retail sales and EGM expenditure would not necessarily indicate that expenditure had switched to EGMs, but it would highlight a relationship that was worth further study. As with our earlier modelling of consumption as a whole, an ARDL specification is used (which allows for the influence of the history of expenditure and income as well as their current values). As we were faced with an uncertain lag structure, and wanted to be able to undertake a Chow test for the year 1995, we needed a series which stretched back at least 8 or
9 years before then. This precluded the use of the series used in the aggregate equation described in section 2.2.1; due to changes in the National Accounts this data are not available prior to 1989-90. Instead we used the components of state final demand (SFD) in the national accounts (ABS 5206, Table 87).
Figure 2.2
Rates of Growth in Categories of Consumption Expenditure
-20 -10 0 10 20 30 40 86-87 88-89 90-91 92-93 94-95 96-97 98-99 00-01 02-03 Year Pe r C e n t
Hotels , cafes , res taurants Other gam bling
Clothing & footwear Rent & other dwelling s ervices EGMsa
Notes: a The first year’s change in EGM expenditure has been excluded as EGMs were not available for all of 1994-95,
and hence the expenditure growth from 1994-95 to 1995-96 is exaggerated. Source: ABS; calculations by the researchers.
Figure 2.3
Rates of Growth in Categories of Consumption Expenditure
-20 -10 0 10 20 30 40 86-87 88-89 90-91 92-93 94-95 96-97 98-99 00-01 02-03 Year P e r Cent
Food Cigarettes & tobacco Alcoholic beverages Recreation & culturea
Notes: a Expenditure on gambling has been subtracted from ‘recreation and culture’ to prevent double counting.
Figure 2.4
Rates of Growth in Categories of Consumption Expenditure
-20 -10 0 10 20 30 40 86-87 88-89 90-91 92-93 94-95 96-97 98-99 00-01 02-03 Year P e r Cent
Electricity, gas & other fuel Health
Education s ervices Ins urance & other financial s erv's Furnis hings & hous ehold equipm ent
Source: ABS; OESR, Queensland Treasury, Australian Gambling Statistics 2005, calculations by the researchers. Figure 2.5
Rates of Growth in Categories of Consumption Expenditure
-20 -10 0 10 20 30 40 86-87 88-89 90-91 92-93 94-95 96-97 98-99 00-01 02-03 Year P e r Cent
Purchas e of vehicles Operation of vehicles Trans port s ervices Com m unications Other goods & s ervices
Source: ABS; OESR, Queensland Treasury, Australian Gambling Statistics 2005, calculations by the researchers.
There is no direct equivalent to ‘household disposable income’ in the SFD series, so we needed to identify an alternative. The two logical alternatives were compensation of employees (wages) and total consumption expenditure. Compensation of employees has the advantage that it is an income measure, but the disadvantage that it does not include non-wage income, nor does it include calls on income such as taxes, so it is not a perfect measure of the resources households have available. Consumption is a better reflection of household’s resources, however, it is obviously not a direct measure of income, and does not necessarily
move in line with changes in income because savings decisions can change. To make the choice the three variables were graphed together from 1989-90. This showed that the growth of compensation of employees over time was very close to that of disposable income, whereas consumption diverged significantly, hence compensation of employees was used in the analysis. Data on gambling expenditure came from Australian Gambling Statistics 2005, produced by the Office of Economic and Statistical Research, Queensland Treasury.
For each of the types of expenditure a simple regression model was estimated with the dependent variable being expenditure on the category of retail sales in question, and the explanatory variables being: compensation of employees in the current period; compensation of employees in previous periods; expenditure on gambling66; a time trend; and previous periods’ expenditures on the category in question. Initially, a maximum of the three previous periods for each of compensation of employees and expenditure on the consumption category in question was included in the model.
These lagged structures were then ‘tested down’ to identify the best system of lags, with lags being removed one at a time, and the explanatory power of this new lag structure being compared to the previous one using tests of model specification (using the Akaike Information Criteria and Schwarz criteria), as well as the value of the Durbin-Watson statistic to ensure that removing the lag hadn’t introduced autocorrelation. In most cases it was only the current period’s income, and only the immediate previous period’s consumption which were significant, but there were a few exceptions.
As this is a relatively generic structure it will generally not produce the best model possible (e.g., it will miss out consumption category specific events, such as the downturn in new car sales in the year from the announcement of the New Tax System and the actual introduction of the GST), but it should be a good approximation. This confidence in the general model structure was borne out by the explanatory power of the estimated models. In each case the F-test indicated that the included variables were jointly very significant in estimating the behaviour of the explanatory variable.
As with our model for total consumption, the impact of gambling expenditure on other forms was tested for in two ways. The direct test for its impact is to include current expenditure on gambling in the ARDL equation. If its coefficient is significantly different from zero this indicates that there is a correlation between the expenditure variable being tested and gambling expenditure. Even if the coefficient for gambling expenditure is significant, this does not necessarily indicate that that gambling expenditure is causing the change in consumption. It could be gambling causing the change in consumption, it could be changes in the consumption expenditure causing changes in gambling, or the changes in both variables could be jointly caused by some other factor which isn’t in the model, such as unanticipated increases in household wealth.
As discussed in the previous section, even if the coefficient for gambling expenditure is not significant it is still possible that gambling has impacted on expenditure through changes in the overall pattern of consumption of this sub-category of expenditure after the legalisation of electronic gaming machines. This can be tested for with the Chow test, which tests whether there has been a significant change in a particular data relationship after a specific point in time. The point of time chosen in this instance was 1994-95, the introduction of electronic gaming machines to hotels and clubs in SA. If the Chow test is significant then this shows
66 Note that we used gambling as a whole rather than EGM and non-EGM gambling, as it is not possible to take the log of zero,
that the factors influencing consumption behaviour were different after that point in time than they were before it.
As there are 17 models, the full results are not presented here; instead Table 2.2 presents the results for the significance of gambling, particularly gambling on electronic gaming machines, in explaining expenditure levels in each of the 17 sub-categories. The first four columns present the coefficient for gambling expenditure estimated in the regression, and information on its statistical significance. The next three columns show the results of the Chow test, and consequently whether there was a break in the series in 1994-95.
Table 2.2
Significance of Gambling Expenditure on Categories of Consumption Expenditure Dependent Variable is Log Expenditure on the Consumption Category
Significance of Gambling variable Chow breakpoint test 1994-95
Expenditure Category
Coefficient Std. Error Prob.
Di fferen t fro m zero ? F-statistic Prob. Di fferen t af te r 19 95 ? Alcoholic beverages 0.057 0.117 0.637 " 2.335 0.137 " Cigarettes & tobacco -0.174 0.130 0.208 " 2.358 0.182 " Clothing & footwear -0.091 0.092 0.341 " 1.591 0.266 "
Communications 0.087 0.095 0.374 " 1.738 0.244 "
Education services 0.010 0.083 0.906 " 0.681 0.621 "
Food 0.099 0.050 0.068 " 0.589 0.710 "
Health -0.010 0.164 0.950 " 1.502 0.290 "
Furnishings & household
equipment -0.129 0.109 0.259 " 3.611 0.053 !
Hotels, cafes and
restaurants 0.201 0.200 0.334 " 4.481 0.038 !
Insurance and other
financial services 0.300** 0.078 0.002 ! 0.389 0.812 " Other goods and services -0.041 0.040 0.318 " 2.223 0.139 " Recreation & culture
(-gambling) -0.018 0.051 0.731 " 0.998 0.453 "
Rent & other dwelling
services -0.003 0.019 0.897 " 1.194 0.485 "
Transport services 0.076 0.084 0.381 " 2.658 0.096 ! Electricity, gas & other fuel 0.087 0.094 0.375 " 0.478 0.783 " Purchase of vehicles 0.142 0.124 0.276 " 0.090 0.991 " Operation of vehicles -0.039 0.055 0.484 " 1.164 0.383 "
Note: * significant at the 1% level ** significant at the 5% level
a. Heteroskedascticity was present in this model, and so it was estimated using White’s Heteroskedascticity Consistent Standard Errors, to ensure that the standard errors were correctly estimated.
The results of the consumption function equations show that gambling only had a statistically significantly impact on one of the categories of consumption expenditure, ‘food’. The coefficient for gambling expenditure in this equation is positive, indicating that as gambling expenditure increases, so does expenditure on food. As this is a log form the coefficient is effectively the elasticity, hence the coefficient suggests that a 1 per cent increase in gambling expenditure leads to a 0.1 per cent increase expenditure on food. The interpretation of this is not straightforward. If there was reason to believe that these two forms of expenditure were complements (e.g., expenditure on coffee and sugar would generally be expected to influence one another as many people will only consume coffee with sugar) then the interpretation would be simple. There does not seem any reason why spending money on gambling would
be complementary with expenditure on food, nor vice versa. This suggests that there may be some exogenous variable influencing expenditure on both ‘food’ and ‘gambling’.
Turning to the results of the Chow test, there were three expenditure categories which had a structural break in their time trend in 1994-95. These were:
• Furnishings and household equipment;
• Hotels, cafes and restaurants; and
• Transport services.
In order to interpret these results it is necessary to estimate the two component models for expenditure on this category; from 1985-86 to 1994-95, and from 1994-95 to 2003-04. It is the values, and significance of, the coefficients (particularly for gambling expenditure) which determine the meaning of this break in the series, and so each of the variables will be discussed in turn.
In the case of ‘furnishings and household equipment’ the coefficient for gambling expenditure is positive and significantly different from zero up to 1994-95, and negative but insignificant from 1994-95 on. The other important differences between the pre- and post- 1994-95 equations are that income has a more significant positive impact on expenditure, and there is a significant negative time trend (which was insignificant in the 1985-86 to 1994-95 equation). These results may suggest that, all other things being equal, gambling expenditure had a relatively more negative impact on ‘furnishings and household equipment’ expenditure after 1994-95 than before, but the other changes are somewhat anomalous.
As a further test for the stability of the model the CUSUM Squared test was run for the full time series. The CUSUM Squared test (Brown, Durbin, and Evans, 1975, quoted in E-views econometrics software) is based on the cumulative sum of the squared recursive residuals. This test plots the cumulative sum together with the relevant 5% critical lines. The test finds parameter or variance instability (and hence a potential structural break in the series) if the cumulative sum goes outside the area between the two critical lines (see Figure 2.6).
In this case there are several points at which the test statistic equals or passes its critical values, indicating potential structural breaks in the series. This means that whilst the introduction of EGMs may (or may not) have changed the pattern of expenditure on ‘furnishings and household equipment’, there have been other factors changing the pattern of expenditure. Therefore it is impossible to isolate the impact of the introduction of EGMs, and any conclusions need to be tentative.
In the case of hotels, cafes and restaurants the impact of gambling expenditure is more straightforward. Prior to 1994-95 gambling had a significant, positive, coefficient; post-1994- 95 it had a negative and significant coefficient.67 The other difference is that prior to 1994-95 expenditure on this category, ceteris parebis, tended to decline with income, whereas after 1994-95 it was increasing with income. The results of the CUSUM Squared test indicate that 1994-95 is the only point of instability in the series.
67 For the pre-1994-95 period, the relevant results for the impact on Hotels, Cafes and Restaurants expenditure of Log Gambling
were Coefficient: 1.516; Std. Error: 0.2445; t-Statistic: 6.198; Prob. equal to zero: 0.0085. Post-1994-95, the results were Coefficient: -0.934; Std. Error: 0.424; t-Statistic: -2.201; Prob. equal to zero: 0.079.
Figure 2.6
Results of the CUSUM Squared Test for the ‘Furnishings and Household Equipment’ Expenditure Equation
The results for the equations before and after 1994-95 could suggest that two factors were at work, firstly that growth in expenditure on gambling after the introduction of EGMs was to some degree due to reductions in expenditures on ‘hotels, cafes and restaurants’, although whether this represents transfers of expenditure within premises which have EGMs (e.g., spending less on food in a hotel as it provides food in its gaming room, or less on alcohol as more money is being spent on EGMs) or a transfer of expenditure between venues that do not and those that do have EGMs (i.e., includes from cafes and restaurants to hotels and from hotel to hotel) cannot be identified with this level of data. In our view, it is most likely to represent a combination of the above, including a transfer from cafes and restaurants to hotels. The second effect, the apparent shift from ‘hotels, cafes and restaurants’ being an inferior