Keywords

1 Background

The construction industry encompasses a variety of activities and is a vital sector in any economy [1, 2]. Construction has a strong linkage with most of the other economic activities of a country [3–5]. It is considered to be an important partner in economic growth and mirrors the stage of economic development [3, 6, 7]. There are opposing views with regard to the relationships between construction and the economy of a country. One view is that construction causes the economy to grow as it creates physical facilities that are needed in the development of other productive activities [2, 8–10]. The opposing view holds that GDP causes the construction output growth [11–13]. Interestingly some other studies suggest a bi-directional relationship between different sub-sectors of the construction industry and the national economy [3, 14–16]. This paper believes that a better understanding of this lead/lag relationship requires more evidence from other countries. It is particularly necessary for developing countries because most of the previous studies used data from developed countries. This paper presents a test on the link between construction and the economy using empirical data from Sri Lanka. It is intended to extend knowledge on causal relationships and other general construction issues related to the case study country, Sri Lanka.

Sri Lanka is a developing country with a per capita income of US$2041 [17]. Sri Lanka has a strong construction sector which contributes significantly to its GDP. The total value of new construction is above 60 % of the total gross domestic fixed capital formation and the industry provides employment to around 7 % of the total labour force [18]. The construction industry in Sri Lanka is expected to grow steadily and systematically in the long run while being a significant and integral part of the economy [4].

2 Description of Methods

The study, on which this paper is based, uses Granger Causality Test to determine the above relationship. Granger Causality Test is an econometric technique pioneered by Granger and Newbold [19] which is applied to find relationships between economic variables. Granger causality uses regression to find the causal relationships between two variables, ‘Xt’ and ‘Yt’ [3, 11, 19]. The regression provides statistical evidence whether the current ‘Y’ value can significantly be explained by the past values of ‘Y’ and ‘X’. Granger causality test considers two autoregressive (AR) models illustrated below:

$$ \mathrm{ Xt}=\sum {\acute{\mathrm{ a}} 0\mathrm{ iYt}-\mathrm{ i}} +\sum {\hat{\mathrm{ a}} 0\mathrm{ iXt}-1+ \mathrm{ ut}} $$
(128.1)
$$ \mathrm{ Yt}=\sum {\acute{\mathrm{ a}} 1\mathrm{ i}\mathrm{ Xt}-\mathrm{ i}} +\sum {\hat{\mathrm{ a}} 1\mathrm{ i}\mathrm{ Yt}-1+\mathrm{ vt}} $$
(128.2)

Where n is the maximum number of lagged observations included in the model and Ut and Vt are the random error terms for each time series. Where causality implies, X is Granger causing Y when á0i is not zero in Eq. 128.1. Similarly, Y is Granger causing X if á1i is not zero in Eq. 128.2. If both of these events occur, then feedback effects exist.

Testing causality involves using F-tests to ascertain whether lagged information on a variable Y provides any statistically significant information about a variable X in the presence of lagged X. If not, then “Y does not Granger-cause X.” There are four possible outcomes in a Granger causality test [20];

  • No causal relationship between two variables,

  • Unidirectional causality from X to Y,

  • Unidirectional causality from Y to X, and

  • Bidirectional causality (X causes Y and Y causes X).

Two sets of indicators are used in the current study. One set represents construction while the second national economy. Construction performance is measured using construction value added (CVA), value of construction in the total gross domestic fixed capital formation (CGDFCF), and construction cost indices (CCI). For the national economy, Gross Domestic Product (GDP), Gross Domestic Product Deflator (GDPD), Unemployment Rate (UE) and Balance of Trade (BT) were considered. Previous studies commonly used GDP as an indicator of economic growth, except [9] who used balance of payment and domestic prices. The indicators used for this study were based on constant prices for the period 1990–2009, and were obtained from published data by the Central Bank of Sri Lanka [18, 20, 21].

3 Results

Time series, especially economic data in level form is non-stationary and most statistical methods including Granger causality require the time series to be transformed to stationarity [19, 22]. Stationary could be detected using three methods: autocorrelation function (ACF), correlogram (Q-statistic), and the Unit Root Test. Among them Unit Root Test is widely used to detect and transform the series into stationary forms [22]. Unit root test provides several tests such as Dickey Fuller (DF), Augmented Dickey Fuller (ADF) and Phillips Perron (PP) for testing the order of integration [23, 24]. For the purpose of this study, Augmented Dickey Fuller (ADF) test is used following recommendations of past studies because the test considers the situation where the white noise error terms are correlated, which is an improvement of Dickey Fuller (DF) test [25]. Table 128.1 presents the critical and calculated values for the selected indicators at 5 % significance level. The comparison of critical values with calculated values reveal that BT and CGDFCF are stationary at first difference. The calculated values fall within rejection region implying that the null hypothesis is rejected, and that the time series has no unit root. Similarly the values for GDP, GDPD, and CVA indicate that they are stationary at second difference. CCI was however found to be non-stationary at second level of difference.

Table 128.1 Unit root test results for variables
Full size table

The regressions (1) and (2) described previously was run for the possible lag values of each variable. The number of lags in causality test is arbitrary. It depends on the relationship between the variables. The causality between variables is described in the following three sections. Each section takes one economic indicator and runs the pair-wise regression with all three indicators for construction. To test causality, the results were validated using the residual plots, auto correlation function (ACF) and Durbin-Watson (DW) statistic. This paper considers the DW statistic method only.

3.1 Causality Between CGFCF, CVA, and CCI and GDP2

The direction of the causality between GDP and CVA was investigated by testing the hypothesis that GDP does not cause CVA and CVA does not cause GDP, which is reported in Table 128.2 (CGFCF,Footnote 1 CVA,Footnote 2 and CCIFootnote 3). Using the probability value of 0.02702, DW statistic of 1.94, it can be concluded that CVA does not cause GDP. Similarly, the causality between GDP and CGDFCF; and between GDP and CCI; were tested and the results are presented in Tables 128.3 and 128.4. Results indicate that CGDFCF does not cause GDP and CCI does not cause GDP for any lag. On the contrary, GDP is found to cause CGDFCF for lags 2 and 4. DW statistics of 1.85 and 1.64 for the latter models shows that there is no serial correlation between the error terms.

Table 128.2 Causality between GDP2 and CVA2
Full size table
Table 128.3 Causality between GDP2 and CGDFCF1
Full size table
Table 128.4 Causality between GDP2 and CCI3
Full size table

3.2 Causality Between GDPD2 and CVA2, CGFCF1, and CCI3

The results of the hypothesis test for causality between GDPD and CVA are given in Table 128.5. It shows that GDPD causes CVA for lag 7 and not vice-versa with a DW statistic value of 1.92. The column 3 in Table 128.6 indicates that the probability for hypothesis GDPD does not cause CGDFCF are insignificant at 5 % level for lags up to 6. However, column 4 indicates that the probabilities for null hypothesis of ‘CGDFCF does not cause GDPD’ are significant for lags up to 5. The respective DW statistics of 1.86 and 1.93 confirms the validity of the models. Therefore, it can be inferred that CGDFCF does cause GDPD and not vice-versa. Table 128.7 shows that probabilities for hypothesis ‘GDPD does not cause CCI’ are significant up to lag 2 while for the null hypothesis ‘CCI does not cause GDPD’ none of them are significant. Thus, it can be inferred that GDPD does cause CCI and not vice-versa. The DW statistic value of 1.87 indicates no auto correlation between residuals.

Table 128.5 Causality between GDPD2 and CVA2
Full size table
Table 128.6 Causality between GDPD2 and CGDFCF1
Full size table
Table 128.7 Causality between GDPD2 and CCI3
Full size table

3.3 Causality Between BT1 and CVA2, CGFCF1, and CCI3

Table 128.8 shows that probabilities for the null hypothesis ‘BT does not cause CCI’ and ‘CCI does not cause BT’ are insignificant at 5 % level for all possible lags except 2. According to Table 128.9 the probabilities for null hypothesis ‘BT does not cause CVA’ and ‘CVA does not cause BT’ are insignificant for all possible lags except 7. Table 128.10 indicates that probability values for null hypothesis ‘BT does not cause CGDFCF’ are not significant at 5 % confidence level up to lag 4 while for null hypothesis ‘CGDFCF does not cause BT’ is insignificant for all possible lags.

Table 128.8 Causality between BT and CCI3
Full size table
Table 128.9 Causality between BT and CVA2
Full size table
Table 128.10 Causality between BT and CGDFCF1
Full size table

4 Conclusions

As it is well documented in construction economics literature, when an economy is booming, the construction sector also booms. Conversely a slowing down in the economy slows down construction activities. Hence, a causal relationship between them could be postulated. What is unknown is which causes what? Such information is useful in policy planning to prioritize investment opportunities. The current study considered the most appropriate among the available indicators to represent both the construction sector and the national economy. A summary of the results obtained from the Granger causality test is given in Table 128.11. For all indicators except CGDFCF and GDPD the cause-effect analysis reveals that the economy leads the construction sector and not vice versa. This supports the viewpoints of [11, 12] that GDP tends to lead construction flow. The results however contradict the views expressed by [2, 10] that construction lead the national economy and that growth in construction precedes growth in GDP.

Table 128.11 Causality between construction and the national economy
Full size table

This finding could be justified for a developing country like Sri Lanka where generally construction is subject to fluctuations. During periods of rapid economic expansion, construction output usually grows faster than those of other sectors but during periods of stagnation the industry is the first to suffer. Government being the major client of the construction industry (contributing nearly two-third of the total annual output in construction) could use the construction sector as an economic regulator whereby it could reduce construction demand by cutting back on construction projects or investment funds when an economy is overheating. This deliberately stimulates investment during periods of unemployment and slack demand. Thus, the Sri Lankan economy could prioritize investment so as to increase economic growth and optimize the use of the construction sector.