4.1. RESULTADOS
4.1.4. ELABORACION DE LA PROPUESTA DIRECTRIZ DE GESTION DE
It is obvious from the results o f the statistical selection that despite the use o f a less restrictive
variable selection procedure, it did not select all o f the theoretical indicators in any case. This is to
be expected because the use of significance levels ensures that only those variables that have met the
specified selection criteria are included. Now, there is a need to re-examine those theoretical
indicators that have not been statistically selected to decide if they should be omitted totally.
Otherwise, they can still be included solely on good theoretical grounds.
For residential construction demand, the indicators that have not been statistically selected are "GDP
deflators", "Interest rate" and "Household formation". Firstly, GDP deflators are indicators of
overall national price changes, covering all sectors namely, consumer, government, investment, and
international components. Although they are considered as the broadest and best indicators of
economy-wide inflation, they may become too general when related specifically to demand in the
residential construction sector. As housing has always been viewed as the single most expensive,
hence important, consumer good of an average household, the Consumer Price Index may be a more
precise indicator o f this type o f demand. This may help to explain why the Consumer Price Index,
and not the GDP deflators, has been selected as one of the significant variables. Secondly, according
to economic theory, interest rate is expected to have a relationship with demand for residential
construction since it is the main determinant o f the cost o f borrowing which affects mortgage loans.
However, this departure from theory may be justifiable in the case o f Singapore as the public sector
has the largest share in the country's residential construction. The ongoing public housing scheme
implemented by the government has to date housed 90 per cent o f the total population. Demand for
government-subsidised housing would not be greatly influenced by the rate o f interest because, in
any case, the substantial savings in purchasing these low-cost housing would have far compensated
formation has often been considered, in theory, as a crucial factor that influences demand for housing
construction. However, quantitative studies (Tang et a l , 1990; and Goh, 1996) have shown the contrary. In this study, the statistical insignificance o f household formation may be attributable to
the use of the national level statistical series "Number o f marriages registered". Again, this may be
a unique case for Singapore as young married couples are encouraged to live with their parents to
form a multi-generation family unit. Hence, for the purpose o f studying housing construction
demand, the number o f marriages registered may not accurately depict household formation in
Singapore. However, no alternative series was available to represent this indicator.
The only indicator that was found to be insignificant for industrial construction demand is "GDP per
head". In general, this indicator reflects national output per person, which acts as a good guide to
a country's general hving standards. Although generally useful, it may bear little direct relations with
the industrial sector. On the other hand, it should have more direct and immediate influence on the
residential and commercial sectors since the general living standards o f a nation usually imply the
population's spending power on items such as housing and retail goods. This line o f reasoning is
supported by GDP per head having been found to be a significant indicator o f both residential and
commercial construction demand.
In the case of demand for commercial building construction, three theoretical indicators were found to be statistically insignificant and they are, "GDP deflators", "Fixed investments" and "Population".
Firstly, GDP deflators may, again be too wide or general in terms o f measuring the level o f inflation
as it covers many other sectors besides the consumer. Since the commercial sector deals mainly with
the retail trade, a narrower indicator of the level o f price change, such as the Consumer Price Index,
may have more direct effects on this consumer-based sector and, hence, the level o f construction
demand. Secondly, the indicator "Fixed investments" has been represented by the national level
statistics "Total cost of commercial developments", which may have measured and reflected only the
cost o f construction. In land-scarce Singapore, the cost o f construction constitutes only a small
proportion o f total development cost which also includes the cost o f land. Hence, decisions to invest
in commercial developments are chiefly influenced by the prevailing price o f commerical land. This
point is substantiated by "Commercial land price" being one o f the significant indicators. Thirdly,
although population size was not found to be significant by itself, this does not mean that it has no
influence on the commercial sector at all. This indicator is often used as a yardstick for minimum
GDP growth because of its use in GDP per head. When interpreting GDP per head, real GDP must
grow faster than the size of population if living standards are to improve. As explained earlier,
and, therefore, buildings to contain such increased activities. From this, it may be suggested that
although the size of population does not relate directly to demand for commercial buildings, it does
have its influence through GDP per head.
Justification can be found for all the unselected indicators, indicating that the statistical approach
has successfully achieved its objective o f sieving out the less suitable variables. However, there are
some general shortcomings o f such statistical procedures which need to be addressed here. Firstly,
it has been emphasized that automated variable selection procedures are not the panacea many users
have been led to expect; prior information about the model can be useful to the selection of a more
suitable set o f modelling variables (Freund and Littell, 1986). From a statistical viewpoint. Draper
and Smith (1981) attributed the shortcomings o f statistical procedures to the lack o f knowledge of
the magnitude of the true random variance o f the observations for any single well-defined problem,
making the choice of a best regression equation impossible. Hence, it becomes necessary to rely on
personal judgement in any o f the selection methods used. Secondly, a major limitation of any
statistical analysis is that the outcome is largely dependent on the quality and quantity o f data used.
For quahty, this is particularly true when dealing with archival data, as opposed to empirical, since
there is less control over the recording and updating o f data; the competence o f the personnel maintaining the data banks has to be relied upon. With regard to quantity, those indicators that were
found to be insignificant may perhaps become significant when more data become available.