This study aims to address the research gap in forecasting output-based Building Cost Index (BCI) for building projects by integrating macroeconomic indicators into a robust econometric model.
Quarterly Australian data from 1997 to 2024 on BCI, gross domestic product (GDP), construction price index, petroleum prices and Producer Price Indexes (PPI) were analysed using Granger causality and Johansen cointegration tests to examine temporal and long-run equilibrium relationships. Based on these results, a Vector Error Correction Model (VECM) was developed and validated.
GDP, petroleum prices and PPIs for electrical, steel and other materials emerged as leading indicators of BCI. Residual diagnostics confirmed no serial correlation and homoscedasticity, validating the VECM assumptions. Out-of-sample forecasts achieved RMSE = 11.48, MAE = 8.54, MAPE = 6.79% and Theil’s U = 0.05, indicating strong predictive accuracy.
This research used Australian data; therefore, its applicability to other regions may be limited. Future studies should replicate the model across diverse geographic and economic contexts to validate its robustness.
The model enables construction managers and policymakers to anticipate cost escalation and plan budgets proactively, particularly during economic volatility. For example, rising petroleum prices or material costs can signal future BCI increases, supporting timely procurement and risk mitigation strategies.
This research introduces a validated VECM tailored to output-based BCI, offering a practical tool for early cost planning and strategic decision-making in building projects.
1. Introduction
The construction industry is a major contributor to economic growth, yet its costs are susceptible to macroeconomic conditions (Wang et al., 2024). To monitor these fluctuations, the Australian Bureau of Statistics (ABS), the national agency responsible for collecting Australia’s official economic and demographic data, publishes the quarterly output-based Building Cost Index (BCI), which reflects the actual prices paid to builders, including labour, materials, overhead and profit margins. Unlike input-based indices that track selected material and labour costs, BCI captures broader market dynamics, making it a critical tool for cost planning (ABS, 2024c).
Construction cost indices fall into two categories: input-based and output-based. Input-based indices, such as the Construction Cost Index (CCI), track changes in selected material and labour costs, offering a narrow view that overlooks overheads and profit margins. In contrast, the output-based BCI, published by ABS, reflects the actual prices paid to builders, including labour, materials, overheads and profit margins. This broader scope makes BCI a more comprehensive indicator of market dynamics and final project costs, which is critical for accurate forecasting and strategic planning.
Macroeconomic indicators, including GDP, CPI, Producer Price Indexes (PPIs) for construction materials and petroleum or crude oil prices, influence BCI by affecting input costs and demand conditions (Kim et al., 2022; Mao et al., 2021; Mao and Xiao, 2019). Although several studies have examined macroeconomic influences on construction cost indices (e.g. Kissi et al., 2019; AlTalhoni et al., 2025; Aslam et al., 2023), most of them focus on input-based indices or highway projects (Shiha et al., 2020; Wang and Ashuri, 2017; Shahandashti and Ashuri, 2016). Research on output-based BCI forecasting for building projects remains scarce, particularly those employing models that capture long-run equilibrium relationships between macroeconomic drivers and output prices. Since the leading indicators vary based on the project types and economic conditions of the specific country, the leading indicators identified for highway and other projects in different countries are not suitable for building projects in Australia.
To create a clearer conceptual link between these empirical gaps and the modelling approach, it is important to understand how macroeconomic shocks influence firms’ pricing behaviour. Output prices respond n to cost pressures but also to firms’ adjustments to changing demand and input conditions. These dynamics underscore the need for a theoretical framework capable of explaining how builders modify prices in both the short run and the long run when economic conditions shift.
To address this gap, this study adopts producer theory as its theoretical foundation. Producer theory models a firm’s profit-maximisation behaviour under changing input costs and output prices, providing a structural rationale for anticipating price adjustments in response to economic conditions (Pindyck and Rubinfeld, 2018). Under this framework, builders adjust output prices to maintain profitability when input costs rise or demand shifts. GDP reflects aggregate demand, influencing firms’ pricing strategies; CPI captures general inflationary pressures; unemployment rate indicates labour market conditions that affect wage costs; petroleum prices represent energy and transport cost shocks; and PPIs track upstream material price volatility. These mechanisms justify the inclusion of GDP, CPI, unemployment rate, petroleum prices and PPIs as explanatory variables. Furthermore, because producer theory implies both short-run adjustments and long-run equilibrium relationships between input costs and output prices, VECM is appropriate for capturing these dynamics (Johansen, 1995; Lütkepohl, 2005). The Vector Error Correction Model (VECM) translates these theoretical mechanisms into an empirical form by considering short-run adjustments alongside a long-run cointegration relationship, thereby providing a direct test of the behavioural dynamics implied by Producer Theory. The objective of this research is to identify leading macroeconomic indicators and develop a validated VECM for forecasting BCI. The model aims to improve cost prediction accuracy and provide actionable insights for construction managers and policymakers to anticipate cost escalation and mitigate risks during economic volatility.
2. Literature review
This section reviews existing research on cost indices and evaluates forecasting approaches that use macroeconomic variables. To support this review, Table 1 summarises key studies, including their purpose, methods, predictors and findings, highlighting the dominance of ARIMA for CCI forecasting and the limited application of VECM for BCI.
Literature review matrix
| Authors | Purpose | Methods | Predictors | Key findings | Output |
|---|---|---|---|---|---|
| AlTalhoni et al. (2025) | Propose highway construction cost index (HCCI) prediction models | VECM, long short-term memory (LSTM) networks and ARIMA | Oil prices, asphalt (PPI), concrete (PPI), hot-finished steel bars (PPI), construction sand, gravel and crushed stone (PPI), number of building permits | VECM is the most effective model for short-term forecasting under volatile conditions, achieving the lowest average mean squared error | HCCI |
| Alzara et al. (2025) | Create a CCI predictive model for Egypt’s construction industry | Long short-term memory (LSTM) and gated recurrent unit (GRU) | PPI, CPI, foreign reserves, oil prices, money supply and EGX30 (Egyptian Exchange) | Oil is the key factor that affects CCI in Egypt | CCI |
| Akpolat (2024) | Examine the impact of real variables on housing prices | Nonlinear autoregressive distributed lag (NARDL) | Exchange rates, mortgage rates, money supply, CCI | The real exchange rate has a positive and symmetric effect | Housing prices |
| Maqsoom et al. (2024) | Analyse the correlation between the construction price index (CPI) or prices of all commodities, lumber and wood products, cement and iron products and inflation in Thailand | ARIMA, Spearman correlation | Inflation rate | Iron products showed a significant relationship with inflation | CPI |
| Al Kailani et al. (2024) | Develop a construction cost index (CCI) for Jordan | Fuzzy analytic hierarchy process (FAHP), ML techniques | Cost of concrete, cement, steel, aggregate, diesel | Random forest model had the lowest MAPE (1.09%) | CCI |
| Zhang et al. (2024) | Observe building cost index (BCI) and CCI reactions concerning labour conditions | Vector autoregression (VAR), granger causality test | BCI, CCI | BCI shows sensitivity to labour supply and unemployment, CCI remains insensitive | BCI, CCI |
| Aslam et al. (2023) | Forecast CCI of building materials in developing countries | ANN, time series, linear regression | Cost of bricks, steel, cement, sand and gravel | ANN model has superior results with the lowest errors | CCI |
| El Said and Stammer (2023) | Identify significant bid items and develop highway construction cost index (HCCI) model | Multiple linear regression, time series analysis | Bid item price | Bituminous material is highly significant for cost estimation | HCCI |
| Hu and Xiao (2022d) | Propose fuzzy cognitive visibility graph (FCVG) for time series forecasting | FCVG, weighted multi-subgraph similarity (WMSS) | M1, M2 and M4 datasets | Leveraging fuzzy interaction improves time-series forecasting | CCI |
| Hu and Xiao (2022a) | Propose multi-subgraph similarity (MSS) for time series forecasting | ARIMA, SARIMA, Holt ES, MSS | CCI | MSS method provides more accurate predictions | CCI |
| Hu and Xiao (2022c) | Propose a forecasting method based on a directed visibility graph | ARIMA, SARIMA, Holt ES, Holt-Winter ES, visibility graph | CCI, GDP | The proposed method offers robust and accurate predictions | CCI, GDP |
| Hu and Xiao (2022b) | Propose a model using recurrent neural network (RNN) and network self-attention | SES, TBATS, ARIMA, CatBoost, transformer, RNN | CCI, M1, M3 data sets | The proposed method performs better for certain time series and shows robustness | CCI |
| Jiang et al. (2022) | Analyse construction costs using multivariate models | SMA, ARIMA, Holt ES | CPI, unemployment rate, employment rate, PPI, crude oil prices, GDP, building permits, import price index, money supply | ARIMA is the best forecasting model with key influencing factors identified | CCI |
| Kim et al. (2022) | Propose a hybrid ARIMA-ANN model for forecasting construction costs | ARIMA, ANN, hybrid ARIMA-ANN | CCI | Hybrid model outperforms individual ARIMA or ANN models for longer-term forecasts | CCI |
| Choi et al. (2021) | Develop city-level CCI models | ARIMA, VECM | CPI, effective federal funds rate, unemployment rate, construction employee ratio, average weekly hours of production, new building permits, M2 money supply, average hourly earnings in construction, S&P 500 stock index, crude oil prices, PPI, housing starts, real personal income, personal consumption expenditures | Significant city-level differences were found; national CCI causes forecast errors | CCI |
| Liu et al. (2021) | Develop a holistic framework for automated HCCI development using inconsistent pay items | Fisher index formula | Pay items | Provides reliable insights into construction market conditions | HCCI |
| Mao et al. (2021) | Forecast CCI using visibility graph network method | Node visibility, network structures | CCI | The method improves prediction accuracy | CCI |
| Cao and Ashuri (2020) | Explore models for volatile cost data prediction | ARIMA, LSTM | HCCI | LSTM outperforms other time-series models | HCCI |
| Zhao et al. (2020) | Propose a time-series transfer function to forecast BCI | ARIMA, time-series transfer function | House price index | Transfer function improves forecasting accuracy by considering time-lag causality | BCI |
| Kissi et al. (2019) | Identify key economic indicators that influence the tender price index prediction in the building industry of Ghana | Mean score ranking, Wilcoxon signed rank | – | The study identifies five significant economic indicators: CPI, PPI, GDP, currency exchange rate and interest rate | – |
| Elem-Uche et al. (2019) | Develop a VECM to examine the relationship between Nigeria’s GDP and key monetary policy variables (credit, exchange rate, and interest rate) | VECM | Money supply, interest rate, domestic credits, real effective exchange rate | Real output depends on money supply, credit and exchange rate channels in the short run and on interest rate in the long run | GDP |
| Elfahham (2019) | Estimate CCI for concrete structures | Neural networks, linear regression, time series | Prices of structural steel, Portland cement, bricks, sand and gravel | The autoregressive time-series prediction method is the most accurate | CCI |
| Mao and Xiao (2019) | Propose a CCI forecasting method using a visibility graph | Visibility graph | CCI | Improves prediction accuracy, contributing to cost-saving in construction | CCI |
| Oteng-Abayie and Dramani (2019) | Examine co-movements between building cost index and inflation and exchange rate in Ghana | Wavelet technique | Non-food CPI, exchange rate | There are co-movements between building cost index and inflation and exchange rate | BCI |
| Zhao et al. (2019) | Forecast residential building costs in New Zealand | Exponential smoothing, ARIMA | BCI | ARIMA outperforms exponential smoothing for townhouses and apartments | BCI |
| Kissi et al. (2018) | Apply autoregressive integrated moving average with exogenous variables (ARIMAX) in modelling tender price index (TPI) in Ghana | ARIMAX, ARIMA | Composite consumer price indices (CCPI), gross domestic product – construction (GDPC), exchange rate (ER), GDP, interest rate and PPI | The ARIMAX model demonstrated better predictive accuracy compared to a single approach such as ARIMA | TPI |
| Moon et al. (2018a, 2018b) | Refine CCI prediction by applying long memory | ARFIMA, ARIMA | CCI | ARFIMA outperforms ARIMA | CCI |
| Moon and Shin (2018a) | Forecast CCI using interrupted time-series model | Interrupted time-series | CCI | The interrupted model performs better than ARIMA and Holt-Winters models | CCI |
| Moon and Shin (2018b) | Propose a CCI forecasting model based on VECM with search query frequencies | VECM | CCI | VECM model shows better predictive ability than cointegrated VAR | CCI |
| Zhang et al. (2018) | Forecast CCI using a visibility graph approach | Visibility graph, | CCI | Proposed method is easier to implement and can forecast CCI with fewer errors | CCI |
| Zhang et al. (2017) | Improve CCI forecasts using fuzzy logic in a visibility graph | Fuzzy logic and visibility graph | CCI | Fuzzy logic improves prediction accuracy by using appropriate rules | CCI |
| Joukar and Nahmens (2016) | Develop predictive model for CCI considering volatility | ARCH, GARCH | CCI | Models show persistent CCI volatility, especially during economic shocks | CCI |
| Shahandashti and Ashuri (2016) | Forecast national highway CCI using multivariate time series models | VECM | CPI, unemployment rate, employment rate in construction, average weekly hours, prime loan rate, PPI, crude oil prices, GDP, GDP implicit price deflator, building permits, construction spending, money supply, average hourly earnings, Dow Jones industrial average, housing starts | Multivariate models are more accurate than univariate ones | HCCI |
| Wang and Ashuri (2017) | Predict CCI using a modified K-nearest neighbours (KNN) algorithm | KNN | CPI, crude oil price, GDP, number of building permits | KNN yields small prediction errors | CCI |
| Cao et al. (2015) | Create a self-adaptive structural radial basis neural network intelligence machine (SSRIM) model to forecast Taiwan CCI | MARS, RBFNN, EL-SVM, GLR, ABC | Wholesale Price Index, CPI, bank lending rates, oil prices, exchange rates, stock indices, Nikkei 225 | Nikkei 225, and bank lending rates are key predictors for Taiwan CCI accuracy | CCI |
| Shahandashti (2014) | Analyse the temporal relationships between highway construction costs and macroeconomic indicators | Pearson correlation, unit root test, granger causality test | PPI, GDP, GDP implicit price deflator, Dow Jones industrial average, money supply, prime loan rate, unemployment rate, federal funds rate, CPI, number of housing starts, number of building permits, construction spending, average hourly earnings, average weekly hours, and employment rate in construction, crude oil price | Crude oil prices and average hourly earnings are leading indicators of highway construction costs | HCCI |
| Cheng et al. (2013) | Establish a hybrid intelligence system (ELSVM) for modelling construction price variations | LS-SVM, DE | Wholesale Price Index, CPI, bank lending rates, oil prices, exchange rates, stock indices, Nikkei 225 | ELSVM model achieves low MAPE in modelling CCI fluctuations | CCI |
| Shahandashti and Ashuri (2013) | Create multivariate time series models for CCI forecasting | Multivariate time series (VEC), granger causality tests | CPI, employment, building permits, money supply, oil price, housing starts | VEC models are suitable for CCI forecasting | CCI |
| Xu and Moon (2013) | Present a VAR model for forecasting the construction cost trend | Cointegrated vector autoregression (VAR) | CCI, CPI | The cointegrated VAR model provides more accurate forecasts | CCI |
| Ashuri et al. (2012) | Identify leading indicators of CCI through empirical tests | Granger causality tests, johansen’s cointegration tests | CPI, oil price, PPI, GDP, employment, building permits, housing starts | Leading indicators include CPI, oil prices, PPI, GDP and employment levels | CCI |
| Ashuri and Lu (2010) | Compare various time series approaches for CCI forecasting | SMA, Holt ES, Holt-Winters ES, ARIMA | CCI | Seasonal ARIMA is the most accurate for in-sample, and Holt-Winters is best for out-of-sample forecasting | CCI |
| Hwang (2009) | Propose dynamic regression models for predicting construction cost index | Linear regression, categorical regression, dynamic regression | CCI | Models generate more objective forecasts and are more accurate than existing models | CCI |
| Williams (1994) | Develop back-propagation neural network (NN) models to predict CCI changes | NN, exponential smoothing and simple linear regression | Housing starts, prime lending rate, CCI changes | Neural networks showed greater error than exponential smoothing and linear regression | CCI |
ABC = artificial bee colony; ARFIMA = autoregressive fractional integrated moving average; ARIMA = autoregressive integrated moving average; CatBoost = categorical boosting; DE = differential evolution; EL-SVM = ensemble learning support vector machine; GLR = generalized likelihood ratio; GARCH = generalized autoregressive conditional heteroscedasticity; Holt ES = Holt’s exponential smoothing; LSTM = long short-term memory; LS-SVM = least squares support vector machine; LS-SVM = least squares support vector machine; MARS = multivariate adaptive regression splines; RBFNN = radial basis function neural network; MA = simple moving average; SARIMA = seasonal autoregressive integrated moving average; SES = simple exponential smoothing; TBATS = trigonometric Box-Cox ARIMA trend seasonal; VECM = vector error correction model; VEC = vector error correction
Extensive research on input-based CCI has used ARIMA and hybrid ARIMA–ANN models, often incorporating macroeconomic indicators including GDP, CPI, petroleum prices and PPIs (Kim et al., 2022; Jiang et al., 2022). These models perform well under stable conditions but assume stationarity and linearity, making them vulnerable to structural breaks and regime shifts. Visibility graph methods and hybrid approaches have emerged to improve accuracy, yet they remain largely descriptive and lack interpretability for policy and managerial decisions.
In contrast, cointegration-based models such as VECM explicitly separate short-run dynamics from long-run equilibrium relationships, offering interpretability and theoretical rigour (Johansen, 1995; Lütkepohl, 2005). Prior applications to construction pricing (Choi et al., 2021; Moon and Shin, 2018b; Shahandashti and Ashuri, 2016) demonstrate VECM’s ability to capture persistent relationships among macroeconomic indicators and construction costs. However, VECM assumes linear adjustment and stable cointegration rank, and results are sensitive to lag selection, limitations that motivate hybridisation with machine-learning techniques for nonlinear dynamics.
Machine-learning models such as ANN, long short-term memory and gated recurrent unit have emerged as alternatives, offering improved accuracy under volatile economic conditions (AlTalhoni et al., 2025; Alzara et al., 2025). While these models handle nonlinearity effectively, they often sacrifice interpretability. They also fail to account for cointegration, which is essential for capturing long-term relationships between macroeconomic indicators and construction costs. This limitation reduces their explanatory power for decision-making, even when predictive performance is strong.
Despite the breadth of existing research, two major limitations remain in the current literature on construction cost forecasting. First, most studies focus on input-based indices (such as CCI), which capture changes in materials and labour inputs but do not reflect the profit margins, overheads and market-driven pricing behaviours embedded in output-based indices. This limits their ability to explain how macroeconomic conditions translate into final construction prices. Second, prior studies generally summarise macroeconomic influences descriptively without grounding variable selection or interpretation in an underlying economic theory of firm behaviour. As a result, the mechanisms through which macroeconomic shocks are transmitted into output prices remain theoretically underdeveloped. In particular, few forecasting studies operationalise Producer Theory to explain how builders adjust output prices in response to input-cost changes and demand fluctuations. This limits the explanatory power of existing models, especially when long-run equilibrium price formation is central to forecasting accuracy. Thus, these limitations highlight the need for a theoretically grounded, equilibrium-based forecasting framework.
This study addresses this gap by integrating Producer Theory with a cointegration-based VECM. The VECM structure directly reflects the theoretical distinction between short-run adjustments and long-run equilibrium among macroeconomic indicators and BCI, enabling a rigorous and interpretable modelling framework for output-based cost forecasting in building projects.
3. Methodology
3.1 Indicator selection and rationale
Five macroeconomic indicators – GDP, petroleum prices, CPI, unemployment rate and PPIs for building materials – were selected based on prior research and producer theory. Producer theory posits that firms adjust output prices to maximise profit under changing input costs and demand conditions (Pindyck and Rubinfeld, 2018). GDP reflects aggregate demand and has repeatedly been shown to influence construction cost indices, as higher economic activity drives demand for building projects (Kissi et al., 2019; Shahandashti and Ashuri, 2016; Hu and Xiao, 2022c). Petroleum prices affect transportation and production costs and are widely recognised as leading indicators in both highway and building cost studies (Moon and Shin, 2018a, 2018b; AlTalhoni et al., 2025; Alzara et al., 2025). CPI reflects general inflationary trends and has been linked to construction cost escalation in multiple studies (Jiang et al., 2022; Choi et al., 2021; Kissi et al., 2019). Unemployment rate captures labour market conditions, which influence labour availability and wage pressures (Zhang et al., 2024; Shahandashti and Ashuri, 2016). PPIs for building materials (e.g. steel, electrical components and concrete) track upstream price movements that propagate to output prices and are supported by studies emphasising material cost volatility (AlTalhoni et al., 2025; Aslam et al., 2023). Collectively, these indicators operationalise producer theory, which posits that firms adjust output prices in response to changes in input costs and broader macroeconomic conditions. By mapping each variable to a specific economic mechanism—such as demand fluctuations (GDP), energy cost shocks (petroleum prices), inflationary pressures (CPI), labour market dynamics (unemployment rate) and material price volatility (PPIs) – the selection process is both theoretically grounded and empirically validated. This alignment reinforces the robustness and interpretability of the forecasting model.
3.2 Data span and sources
Quarterly Australian data from 1997Q3 to 2024Q2 (108 observations) were sourced from ABS catalogues (6427.0, 6345.0, 6401.0, 5206.0, 6202.0 and 6457.0). This period was chosen to capture multiple economic cycles, including pre-pandemic stability, the COVID-19 shock and post-pandemic recovery. These structural breaks and volatility make VECM particularly relevant, as it models both short-run fluctuations and long-run equilibrium relationships – critical for forecasting under persistent macroeconomic shocks.
3.3 Data preparation and transformation
All variables were obtained as indexes except unemployment, which is expressed as a percentage. Before model estimation, stationarity was assessed using the Augmented Dickey-Fuller (ADF) test (Dickey and Fuller, 1979; Gurmu, 2025). Non-stationary series were differenced once to achieve stationarity.
3.4 Econometric framework and parameter selection
The VECM was selected because Johansen cointegration tests (Johansen, 1995) confirmed long-run relationships among BCI and the chosen indicators. VECM captures both short-run dynamics and long-run equilibrium adjustments, making it suitable for contexts with structural breaks and persistent shocks (Lütkepohl, 2005). Lag length was determined using the Akaike Information Criterion (AIC) to balance model fit and complexity, minimising overfitting risk (Cavanaugh and Neath, 2019). Granger causality tests (Granger, 1969) were applied using a 10% significance level, appropriate for exploratory studies with moderate sample sizes and consistent with prior construction management research (Shahandashti and Ashuri, 2016).
3.5 Diagnostic tests and validation
Model diagnostics were conducted to ensure reliability. The VEC Residual Serial Correlation LM Test was conducted to detect any autocorrelation in the residuals, and tests for heteroskedasticity were carried out to verify constant variance (Engle and Granger, 2015). Predictive accuracy was evaluated using RMSE, MAE, MAPE and Theil’s U-statistic on a hold-out sample (2019Q1–2024Q2). These metrics provide a robust assessment of out-of-sample performance (Lima et al., 2023), ensuring the model’s practical applicability for forecasting BCI under real-world conditions.
4. Findings
4.1 Descriptive analysis
Petroleum prices show the highest volatility, while GDP and CPI fluctuate moderately, and key material PPIs vary by category. These broad patterns provide initial context but do not capture dynamic interactions among variables; therefore, the analysis proceeds directly to stationarity testing, causality and cointegration to identify leading indicators.
4.2 Stationarity results
The ADF test was applied to the BCI and all selected macroeconomic indicators to verify stationarity. Table 2 presents the ADF test results for each variable, listing the optimal lag length and p-values. The results indicate that BCI is not stationary in its original form (p-value = 0.79), indicating that the null hypothesis of non-stationarity cannot be rejected. Similar non-stationarity is observed in all macroeconomic indicators in their level forms (Table 2). This lack of stationarity can lead to misleading regression results because apparent relationships may reflect shared trends rather than true causality, reducing the reliability of forecasts. To address this issue, first differencing was applied to each data series. Differencing is a standard approach for removing stochastic trends and stabilising the mean, which is necessary for valid time series modelling and subsequent cointegration analysis. After differencing once, all series achieved stationarity, as indicated by lower p-values (Table 3). This transformation ensures that the data meet the stationarity requirement, thereby permitting the use of multivariate time series tests in subsequent analyses.
Augmented Dickey-fuller (ADF) test results before differencing
| Series | p-value | Lag | Max. lag | Observations |
|---|---|---|---|---|
| BCI | 0.7913 | 1 | 4 | 84 |
| CPI | 0.9483 | 0 | 4 | 85 |
| GDP | 0.8914 | 2 | 4 | 83 |
| PP | 0.3077 | 2 | 4 | 83 |
| PPI__APPLIANCES_ | 0.9912 | 0 | 4 | 85 |
| PPI__CERAMIC_ | 0.9999 | 0 | 4 | 85 |
| PPI__CONCRETE_ | 0.8475 | 4 | 4 | 81 |
| PPI__ELECTRICAL_ | 0.9999 | 0 | 4 | 85 |
| PPI__OTHER_METALS_ | 0.9486 | 1 | 4 | 84 |
| PPI__OTHER_MATERIALS_ | 0.9936 | 0 | 4 | 85 |
| PPI__PLUMBING_ | 0.9922 | 0 | 4 | 85 |
| PPI__STEEL_ | 0.7622 | 3 | 4 | 82 |
| PPI__TIMBER_ | 0.9640 | 1 | 4 | 84 |
| UR | 0.0964 | 4 | 4 | 81 |
Legend: CPI = consumer price index, GDP = gross domestic product, LP = labour productivity, PP = petroleum price, PPI = producer price index, UR = unemployment rate
Augmented Dickey-Fuller (ADF) test results after first differencing
| Series | p-value | Lag | Max. lag | Observations |
|---|---|---|---|---|
| D(BCI) | 0.0005 | 0 | 4 | 84 |
| D(CPI) | 0.0000 | 0 | 4 | 84 |
| D(GDP) | 0.0000 | 1 | 4 | 83 |
| D(PP) | 0.0000 | 1 | 4 | 83 |
| D(PPI__APPLIANCES_) | 0.0000 | 0 | 4 | 84 |
| D(PPI__CERAMIC_) | 0.0000 | 0 | 4 | 84 |
| D(PPI__CONCRETE_) | 0.0020 | 4 | 4 | 80 |
| D(PPI__ELECTRICAL_) | 0.0157 | 4 | 4 | 80 |
| D(PPI__OTHER_METALS_) | 0.0000 | 0 | 4 | 84 |
| D(PPI__OTHER_MATERIALS_) | 0.0000 | 0 | 4 | 84 |
| D(PPI__PLUMBING_) | 0.0000 | 0 | 4 | 84 |
| D(PPI__STEEL_) | 0.0000 | 2 | 4 | 82 |
| D(PPI__TIMBER_) | 0.0000 | 0 | 4 | 84 |
| D(UR) | 0.0050 | 3 | 4 | 81 |
Legend: CPI = consumer price index, GDP = gross domestic product, LP = labour productivity, PP = petroleum price, PPI = producer price index, UR = unemployment rate
4.3 Granger causality
Following the AIC-based lag selection, the optimum lag length was found to be 4 and it is used in subsequent analyses. Although the AIC results show an optimum lag length of 4, Granger analysis was carried out by including additional lag lengths (1, 2 and 3) to check the robustness of the findings. The results show GDP consistently Granger causes BCI at lags 2–4 (p = 0.01 at lag 2, p = 0.02 at lag 3 and p = 0.04 at lag 4) (Table 4). This suggests that past values of GDP have strong predictive power for future changes in BCI, making GDP a significant leading indicator.
Pairwise granger causality tests
| Null hypothesis | Lag 1 | Lag 2 | Lag 3 | Lag 4 | ||||
|---|---|---|---|---|---|---|---|---|
| F-Statistic | p-value | F-Statistic | p-value | F-Statistic | p-value | F-Statistic | p-value | |
| D(CPI) does not granger cause D(BCI) | 0.12 | 0.73 | 1.70 | 0.19 | 2.02 | 0.12 | 1.61 | 0.18 |
| D(BCI) does not granger cause D(CPI) | 0.38 | 0.54 | 0.45 | 0.64 | 0.20 | 0.90 | 0.16 | 0.96 |
| D(GDP) does not granger cause D(BCI) | 0.44 | 0.51 | 4.54 | 0.01 | 3.47 | 0.02 | 2.56 | 0.04 |
| D(BCI) does not granger cause D(GDP) | 8.27 | 0.01 | 6.34 | 0.00 | 4.15 | 0.01 | 3.04 | 0.02 |
| D(PP) does not granger cause D(BCI) | 1.13 | 0.29 | 3.57 | 0.03 | 2.67 | 0.05 | 2.29 | 0.06 |
| D(BCI) does not granger cause D(PP) | 0.01 | 0.91 | 0.06 | 0.94 | 0.11 | 0.96 | 0.46 | 0.77 |
| D(PPI__APPLIANCES_) does not granger cause D(BCI) | 0.03 | 0.87 | 0.02 | 0.98 | 0.25 | 0.86 | 0.18 | 0.95 |
| D(BCI) does not granger cause D(PPI__APPLIANCES_) | 0.00 | 0.99 | 1.10 | 0.34 | 0.68 | 0.56 | 0.81 | 0.53 |
| D(PPI__CERAMIC_) does not granger cause D(BCI)) | 4.11 | 0.05 | 2.87 | 0.06 | 2.23 | 0.09 | 1.79 | 0.14 |
| D(BCI) does not granger cause D(PPI__CERAMIC_) | 0.01 | 0.92 | 0.44 | 0.65 | 1.03 | 0.38 | 2.27 | 0.07 |
| D(PPI__CONCRETE_) does not granger cause D(BCI) | 2.93 | 0.09 | 1.81 | 0.17 | 1.49 | 0.22 | 1.94 | 0.11 |
| D(BCI) does not granger cause D(PPI__CONCRETE_) | 4.54 | 0.04 | 2.16 | 0.12 | 1.89 | 0.14 | 1.31 | 0.27 |
| D(PPI__ELECTRICAL_) does not granger cause D(BCI) | 5.68 | 0.02 | 4.73 | 0.01 | 3.03 | 0.03 | 2.25 | 0.07 |
| D(BCI) does not granger cause D(PPI__ELECTRICAL_) | 0.20 | 0.66 | 0.47 | 0.63 | 0.13 | 0.94 | 0.32 | 0.86 |
| D(PPI__OTHER_MATERIALS_) does not granger cause D(BCI) | 6.55 | 0.01 | 3.21 | 0.05 | 2.34 | 0.08 | 2.14 | 0.08 |
| D(BCI) does not granger cause D(PPI__OTHER_MATERIALS_) | 0.40 | 0.53 | 0.17 | 0.84 | 1.37 | 0.26 | 1.44 | 0.23 |
| D(PPI__OTHER_METALS_) does not granger cause D(BCI) | 1.79 | 0.18 | 1.08 | 0.34 | 1.04 | 0.38 | 0.98 | 0.43 |
| D(BCI) does not granger cause D(PPI__OTHER_METALS_) | 4.18 | 0.04 | 3.64 | 0.03 | 2.63 | 0.06 | 2.46 | 0.05 |
| D(PPI__PLUMBING_) does not granger cause D(BCI) | 0.12 | 0.73 | 0.45 | 0.64 | 0.49 | 0.69 | 0.61 | 0.66 |
| D(BCI) does not granger cause D(PPI__PLUMBING_) | 0.67 | 0.42 | 1.23 | 0.30 | 0.99 | 0.40 | 0.91 | 0.46 |
| D(PPI__STEEL_) does not granger cause D(BCI) | 9.76 | 0.00 | 6.50 | 0.00 | 5.53 | 0.00 | 3.88 | 0.01 |
| D(BCI) does not granger cause D(PPI__STEEL_) | 19.19 | 0.00 | 7.87 | 0.00 | 4.62 | 0.01 | 3.58 | 0.01 |
| D(PPI__TIMBER_) does not granger cause D(BCI) | 0.01 | 0.92 | 1.38 | 0.26 | 1.15 | 0.33 | 1.04 | 0.39 |
| D(BCI) does not granger cause D(PPI__TIMBER_) | 1.49 | 0.23 | 0.82 | 0.44 | 0.43 | 0.73 | 1.01 | 0.41 |
| D(UR) does not granger cause D(BCI) | 1.39 | 0.24 | 0.87 | 0.42 | 0.93 | 0.43 | 0.95 | 0.44 |
| D(BCI) does not granger cause D(UR) | 11.03 | 0.00 | 6.37 | 0.00 | 4.72 | 0.00 | 2.76 | 0.03 |
lag = 4; Sample = 9 / 01/1997–12 / 01/2018; BCI = building cost index; CPI = consumer price index; GDP = gross domestic product; PP = petroleum price; UR = unemployment rate; PPI = producer price index
Likewise, petroleum price consistently Granger causes BCI at lags 2–4 (p = 0.03 at lag 2, p = 0.05 at lag 3 and p = 0.06 at lag 4). PPI for electrical materials also consistently Granger causes BCI at lags 1–4 (p = 0.02 at lag 1, p = 0.01 at lag 2, p = 0.03 at lag 3 and p = 0.07 at lag 4). Similarly, PPI for other materials consistently Granger causes BCI at lags 1–4 (p = 0.01 at lag 1, p = 0.05 at lag 2, p = 0.08 at lag 3 and p = 0.08 at lag 4). PPI for steel consistently Granger causes BCI at lags 1–4 (p = 0.00 at lag 1, p = 0.00 at lag 2, p = 0.00 at lag 3 and p = 0.01 at lag 4). These findings show that GDP, petroleum prices, along with the PPIs for steel, electrical and other materials are the leading indicators for BCI, which can be valuable for forecasting building costs. However, other variables, including unemployment rate, CPI and PPI of some materials do not show significant causality with BCI (Table 4), indicating limited predictive value for these indicators.
4.4 Cointegration results
Table 5 presents the results of the Johansen cointegration test using the trace statistic, which helps determine the number of cointegrating relationships among the variables in the model. The test begins by assessing the hypothesis of “None” (no cointegration among variables) and continues incrementally up to “At most 5” cointegrating equations. For each hypothesised number of cointegrating relationships, the test calculates an eigenvalue and a trace statistic, which is then compared to a critical value at the 5% significance level. If the trace statistic exceeds the critical value, the null hypothesis of that rank is rejected, suggesting the existence of at least one additional cointegrating relationship beyond the specified level.
Results of Johansen cointegration test
| Hypothesized No. of CE(s) | Eigenvalue | Trace statistic | 0.05 Critical value | p-value |
|---|---|---|---|---|
| None* | 0.500731 | 122.3966 | 95.75366 | 0.0002 |
| At most 1 | 0.273236 | 66.13317 | 69.81889 | 0.0950 |
| At most 2 | 0.206013 | 40.28173 | 47.85613 | 0.2126 |
| At most 3 | 0.108350 | 21.59601 | 29.79707 | 0.3215 |
| At most 4 | 0.086802 | 12.30678 | 15.49471 | 0.1428 |
| At most 5* | 0.059302 | 4.951761 | 3.841465 | 0.0261 |
Trace test indicates 1 cointegrating equation (CE) at the 0.05 level, *denotes rejection of the hypothesis at the 0.05 level; lag intervals 1–4
As shown in Table 5, when testing the null hypothesis of no cointegrating relationships (“None”), the trace statistic is 122.40, which exceeds the critical value of 95.75, indicating that the null hypothesis can be rejected. The p -value (0.0002) is also close to zero, providing strong statistical evidence against the null hypothesis. This result suggests that there is one cointegrating relationship among the variables, indicating that they share a long-term equilibrium relationship. This implies that the variables move together over time and deviations from this equilibrium are meaningful rather than random. The presence of cointegration justifies the use of a VECM, which incorporates both short-run dynamics and an error-correction term to capture adjustments towards the long-run equilibrium. Given this finding, a VECM is suitable for further analysis, as it can capture both the short-term dynamics and long-term equilibrium adjustments within a cointegrated system (Lütkepohl, 2005).
4.5 Vector error correction model
The results of the VECM are presented in Table 6. Among the independent variables, the PPI for other materials has the most significant impact on BCI, with a coefficient of −1.66 and a significant t-statistic of −8.19. This suggests that an increase in the price index of other materials leads to a notable long-term decrease in BCI. Similarly, the PPI for electrical materials has a positive and significant influence on BCI, with a coefficient of 0.41 and a t-statistic of 3.78, indicating that rising electrical material prices are associated with higher BCI values over time.
Results of VECM
| Parameters | CointEq1 |
|---|---|
| Long-run coefficients from the VECM cointegration equation | |
| BCI(−1) | 1.000000 |
| PP(−1) | −0.149451 |
| (0.02838) | |
| [−5.26532] | |
| GDP(−1) | 0.152089 |
| (0.21344) | |
| [ 0.71257] | |
| PPI__OTHER_MATERIALS_(−1) | −1.664077 |
| (0.20320) | |
| [−8.18934] | |
| PPI__STEEL_(−1) | −0.043404 |
| (0.04007) | |
| [−1.08324] | |
| PPI__ELECTRICAL_(−1) | 0.405031 |
| (0.10723) | |
| [ 3.77726] | |
| C | 31.74269 |
| Short-run coefficients and adjustment terms from the VECM | |||||
|---|---|---|---|---|---|
| Parameters | Coefficient | Std. error | t-statistic | Prob. | |
| CointEq1 | C(1) | 0.025692 | 0.033126 | 0.775582 | 0.4409 |
| D(BCI(−1)) | C(2) | 0.626100 | 0.131251 | 4.770234 | 0.0000 |
| D(BCI(−2)) | C(3) | 0.122525 | 0.156938 | 0.780722 | 0.4379 |
| D(BCI(−3)) | C(4) | 0.062580 | 0.143548 | 0.435952 | 0.6644 |
| D(PP(−1)) | C(5) | 0.114233 | 0.093985 | 1.215444 | 0.2288 |
| D(PP(−2)) | C(6) | −0.184924 | 0.097998 | −1.887018 | 0.0638 |
| D(PP(−3)) | C(7) | −0.032006 | 0.095435 | −0.335367 | 0.7385 |
| D(GDP(−1)) | C(8) | 0.005308 | 0.006988 | 0.759648 | 0.4503 |
| D(GDP(−2)) | C(9) | −0.012265 | 0.007165 | −1.711789 | 0.0919 |
| D(GDP(−3)) | C(10) | 0.004202 | 0.007491 | 0.560904 | 0.5769 |
| D(PPI__OTHER_MATERIALS_(−1)) | C(11) | 0.050684 | 0.030430 | 1.665567 | 0.1008 |
| D(PPI__OTHER_MATERIALS_(−2)) | C(12) | −0.016066 | 0.032042 | −0.501402 | 0.6179 |
| D(PPI__OTHER_MATERIALS_(−3)) | C(13) | −0.033028 | 0.033041 | −0.999606 | 0.3214 |
| D(PPI__STEEL_(−1)) | C(14) | −0.177731 | 0.096614 | −1.839593 | 0.0706 |
| D(PPI__STEEL_(−2)) | C(15) | 0.000338 | 0.101362 | 0.003334 | 0.9974 |
| D(PPI__STEEL_(−3)) | C(16) | 0.088169 | 0.091008 | 0.968808 | 0.3364 |
| D(PPI__ELECTRICAL_(−1)) | C(17) | −0.056700 | 0.024282 | −2.335016 | 0.0228 |
| D(PPI__ELECTRICAL_(−2)) | C(18) | 0.037898 | 0.029754 | 1.273693 | 0.2075 |
| D(PPI__ELECTRICAL_(−3)) | C(19) | −0.027490 | 0.023437 | −1.172933 | 0.2453 |
| Constant | C(20) | 0.240933 | 0.135195 | 1.782110 | 0.0796 |
Standard errors in (); t-statistics in []; lag = 3
The petroleum price also plays a significant role in the long-run relationship, with a coefficient of −0.15 and a t-statistic of −5.27. This negative relationship suggests that an increase in petroleum prices is associated with a decrease in BCI. This may suggest dominant demand destruction or substitution effects over cost-push inflation in the long run. The PPI for steel has a coefficient of −0.04; however, its t-statistic of −1.08 implies that it is not statistically significant in the long-run relationship. GDP has a positive coefficient of 0.15 but a low t-statistic of 0.71, indicating that higher economic activity increases demand for construction, which could drive up building costs. However, the low t-statistic suggests this relationship may not be statistically significant, and GDP may not have a strong long-term impact on BCI.
In Table 6, the short-run VECM results are presented which provide insights into how various factors influence BCI in the short term. The error correction term (CointEq1) has a coefficient of 0.026 but is statistically insignificant (p = 0.44), indicating that BCI does not significantly adjust back to its long-run equilibrium in the short run. This suggests that short-term variations in BCI are driven more by immediate changes in market conditions rather than the long-term equilibrium relationship. Past values of BCI have a notable impact, with D(BCI(−1)) showing a strong and significant positive effect (coefficient = 0.63, p = 0.00 < 0.05), while D(BCI(−2)) and D(BCI(−3)) are insignificant. This indicates that recent changes in BCI tend to persist, reinforcing the short-term momentum in construction costs. However, the effects of petroleum prices and GDP on BCI are mixed. D(PP(−1)) is insignificant, while D(PP(−2)) has a negative effect (coefficient = −0.18, p = 0.06) at the 10% significance level. D(GDP(−1)) is insignificant, whereas D(GDP(−2)) has a negative effect (coefficient = −0.01, p = 0.09) at the 10% significance level. This suggests that petroleum price changes and the influence of economic activity may take time to influence BCI, possibly reflecting delayed cost adjustments in the construction industry. PPI for other materials shows a borderline positive effect in the short run, with D(PPI_OTHER_MATERIALS (−1)) having a coefficient of 0.05 (p = 0.10). While this aligns with expectations that higher material costs drive up construction prices, the effect is weak and inconsistent across different lag periods.
D(PPI_STEEL(−1)) has a borderline negative effect (coefficient = −0.18, p = 0.07), and D(PPI_ELECTRICAL(−1)) has a significant negative effect (coefficient = −0.06, p = 0.02), suggesting that an increase in electrical material costs leads to a decline in BCI. The constant term (coefficient = 0.24, p = 0.08) is borderline significant, suggesting an underlying upward trend in BCI.
4.6 Diagnosis test results
Table 7 presents diagnostic tests on the residuals of the VECM to evaluate the model’s validity. According to the test, no significant serial correlation was observed at all lags (p = 0.39 at Lag 1, p = 0.49 at Lag 2, p = 0.39 at Lag 3). Thus, the null hypotheses that state residuals are uncorrelated cannot be rejected. This indicates that the residuals are not autocorrelated, suggesting that the model captures the data’s dynamic structure well.
Results of diagnostic tests and model forecast evaluation
| Lag | LRE* stat | df | Prob. | Rao F-stat | df | p-value |
|---|---|---|---|---|---|---|
| VEC residual serial correlation LM test | ||||||
| 1 | 37.73532 | 36 | 0.3899 | 1.054627 | (36, 226.7) | 0.3929 |
| 2 | 35.61494 | 36 | 0.4868 | 0.990938 | (36, 226.7) | 0.4897 |
| 3 | 37.84730 | 36 | 0.3850 | 1.058006 | (36, 226.7) | 0.3880 |
| VEC residual heteroskedasticity test | ||
|---|---|---|
| Chi-sq | df | p-value |
| 906.5849 | 798 | 0.4423 |
| Model forecast evaluation | |||||
|---|---|---|---|---|---|
| Variable | Included observations | RMSE | MAE | MAPE | Theil |
| BCI | 22 | 11.47653 | 8.536784 | 6.786502 | 0.045473 |
Sample: 3 / 01/2019–6 / 01/2024; RMSE: root mean square error, MAE: mean absolute error, Mean Absolute Percentage Error, Theil: Theil inequality coefficient
Similarly, the VEC residual heteroscedasticity test reveals no significant evidence of heteroscedasticity (Table 7). The chi-squared test statistic with a p -value of 0.44 shows that the null hypothesis of constant variance cannot be rejected, indicating stable residual variance over time. These diagnostic tests confirm that the VEC model satisfies the assumptions of no serial correlation and homoscedasticity, strengthening its suitability for accurate forecasting.
4.7 Model validation results
Table 7 presents four key performance metrics: root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and Theil’s U-statistic. These metrics provide insights into the model’s precision. Overall, the VECM performs well in predicting BCI. The model exhibits relatively low RMSE (11.47), MAE (8.54) and MAPE (6.79%) values, indicating accurate forecasts. Theil’s statistic (0.05) value is also close to zero, suggesting a high predictive accuracy and minimal forecasting error. These results confirm that the model provides accurate predictions, supporting its practical application in early cost planning and escalation. Finally, after validating the model, the VECM equation is constructed as follows based on the coefficients shown in Table 6, reflecting the impact of each lagged variable on the Building Cost Index:
where D(BCI) = change in the building cost index; C (1) to C (20) are coefficients indicated in Table 6; GDP = gross domestic product; PP = petroleum price; PPIE = producer price index of electrical materials; PPIO = producer price index of other materials; PPIS = producer price index of steel.
5. Discussion
This study identifies GDP, petroleum prices and the PPIs for specific materials such as steel, electrical and other materials that comprise insulation and plaster products, as the leading indicators for an output-based BCI. In comparison, Shahandashti and Ashuri (2016) identified crude oil prices and average hourly earnings as the primary leading indicators for highway projects. This emphasis on fuel and labour costs aligns with the unique nature of highway construction, where fuel-intensive equipment and specialised labour represent significant cost factors. For a broader construction context, Ashuri et al. (2012) examined the input-based CCI and identified a broader set of leading indicators. These included the consumer price index, crude oil price, PPI, GDP, construction employment levels, building permits, housing starts and money supply. This set of indicators reflects a focus on input availability and sector-specific demand metrics, which heavily influence construction material and labour supply costs. Interpreted through Producer Theory, these contrasts show that although both BCI and HCCI are output-based and therefore reflect firms’ profit-maximising pricing behaviour, the specific leading indicators differ because project types exhibit distinct input mixes, substitution possibilities and cost-transmission mechanisms. In contrast, input-based indices such as CCI respond primarily to changes in the cost of underlying materials and labour rather than to firms’ strategic output-pricing decisions.
Notably, crude oil or petroleum prices emerge as a consistent leading indicator across all studies, whether focused on general construction (Ashuri et al., 2012), building projects (this study) or highway construction (Shahandashti and Ashuri, 2016). This consistency highlights crude oil prices’ widespread influence on construction costs (Olatunji, 2010) through their effects on fuel expenses during transportation and the production costs of numerous construction materials. Variations in oil prices can drive costs for machinery operation, material production and logistics (Gohari et al., 2018) – factors central to all types of construction (Olatunji, 2010). This universal impact explains the inclusion of crude oil prices in both input-based and output-based indices, underscoring the factor’s relevance for both material procurement and final project costs.
The similarities across these studies, such as the consistent use of GDP, underscore the influence of general economic conditions on construction costs (Wang et al., 2022). GDP reflects economic growth, which drives demand for construction. However, the differences in leading indicators reflect the distinct characteristics of output-based versus input-based indices and project-specific versus general construction contexts. This study focuses on output-based indices that track final building project costs, making it sensitive to factors like key materials that directly affect project budgets. Conversely, Ashuri et al. (2012)’s input-based index includes indicators relevant to the availability and cost of construction inputs, such as building permits, housing starts and construction employment levels. These indicators provide insight into sector-specific demand and labour availability, which more directly impact input costs than final project expenses. Such distinctions are consistent with Producer Theory, which emphasises that firms’ pricing decisions depend on output-market conditions and profit optimisation strategies rather than simply on input availability.
Some counterintuitive findings were obtained in this study. For instance, in the long run, PPI for “other materials” has a negative relationship with BCI. This could be due to the substitution effect (e.g. builders switching to cheaper materials) to reduce building costs in response to higher prices of “other materials” such as insulation. The substitution effect suggests that an increase (decrease) in the price of a specific building material, independent of income effects, leads to a decrease (increase) in its demand as consumers shift towards more cost-effective alternatives (Becchetti et al., 2019). This substitution response aligns directly with Producer Theory, which predicts that firms re-optimise their input mix to maintain profitability when faced with sustained increases in certain material costs.
Similarly, in the long run, petroleum price and PPI for steel have negative effects on BCI, this could happen if higher petroleum prices reduce demand for construction, leading to lower building costs. According to Lepage (2025), the rise of oil prices decreases the levels of economic activity and construction firms are discouraged from taking on new projects which will ultimately lower employment rates, productivity and profitability. Likewise, PPI for steel has a negative effect on BCI which suggests that higher steel prices could reduce demand for steel-intensive construction, leading to lower building costs. These adaptive behaviours align with producer theory, which posits that firms maximise profit not only by adjusting output prices but also by altering resource allocation and project specifications under cost pressures. Sustained increases in steel prices may also lead builders to substitute alternative materials or adopt design optimisation strategies to minimise steel usage, as demonstrated by Moynihan and Allwood (2014). These patterns underscore that firms respond to cost shocks not only through price adjustments but also by altering material choices and project scope, consistent with producer theory’s profit-maximisation principle.
In the short run, it was found that steel has negative impacts on BCI, which is unexpected since higher steel prices should typically increase construction costs. This could indicate a shift in material usage or substitution effects or change in project specification in response to price changes (Becchetti et al., 2019). The steel usage in buildings could also be minimised by design optimisation. For instance, Moynihan and Allwood (2014) concluded that steel usage in buildings could be drastically reduced by designing for minimum steel materials. Likewise, in the short term, electrical materials have negative impacts on BCI. This could imply that firms adjust project specifications or delay projects when electrical materials costs rise. Furthermore, the insignificant error correction term suggests that short-run BCI movements are driven more by immediate market shocks and momentum than by rapid adjustment to long-run equilibrium, reinforcing the importance of short-term volatility management in construction planning. This finding indicates that while long-run equilibrium relationships exist, short-run adjustments are weak, emphasising the need for forecasting models that incorporate both immediate market shocks and structural trends, consistent with producer theory’s view of adaptive firm behaviour under cost pressures.
5.1 Theoretical implications
This study deepens the theoretical insights into producer theory by demonstrating how building construction firms adjust pricing strategies in response to macroeconomic conditions. While prior research has often focused on input costs, such as materials and labour, from an input-based CCI perspective (Al Kailani et al., 2024; Hu and Xiao, 2022c; Choi et al., 2021), this study showcases how output-based BCI, provide a more comprehensive understanding of pricing adjustments. By incorporating final project costs, including profit margins and overhead, this model underscores the relevance of broader economic factors in building construction firms’ decision-making, aligning with producer theory’s focus on profit optimisation amid cost pressures (Pindyck and Rubinfeld, 2018). Specifically, GDP reflects demand-side pressures, petroleum prices represent energy cost shocks and PPIs for steel, electrical and other materials capture input cost volatility – each aligning with producer theory’s premise that firms adjust output prices and resource allocation under changing economic conditions. Importantly, the model’s identified 2–4 quarter lag between petroleum price movements and BCI changes provides firms with a clear window to hedge fuel-intensive activities or renegotiate logistics and transport contracts before cost pressures materialise.
By identifying GDP and other indicators as significant predictors, this research demonstrates that output-based indices can capture the complex economic dynamics impacting building construction costs. This output-focused perspective offers an alternative approach that integrates full project costs, adding a new dimension to cost forecasting and supporting the hypothesis that economic growth and inflation are central drivers of building construction cost dynamics. However, findings such as the negative long-run relationship between petroleum prices and BCI challenge simplistic cost-pass-through assumptions, suggesting adaptive behaviours like demand contraction and material substitution, which extend producer theory’s explanatory scope.
In addition, by distinguishing leading indicators for the BCI from those used in highway and general construction indices, this study reveals the unique cost drivers specific to building projects. While crude oil prices affect both building and highway construction, other indicators, such as the PPI for steel and electrical materials, are especially relevant to building construction. This specificity underscores that project type significantly influences which macroeconomic factors should be considered when forecasting costs, offering a refined approach to understanding construction cost fluctuations. This research signals a paradigm shift from generic, input-based cost forecasting towards project-specific, macro-integrated models that capture long-run economic relationships, bridging theory and practice for more accurate and context-sensitive predictions.
5.2 Practical implications
The development of the BCI forecasting model has practical implications for builders, developers and policymakers. By providing an output-based perspective that integrates key economic variables, the model offers a tool for more accurate budgeting and risk management. Building construction managers and other building practitioners can use this model to anticipate cost escalations linked to macroeconomic shifts, allowing them to make more informed decisions about project scheduling, budgeting and resource allocation. For example, a developer planning a high-rise building project during a period of rising petroleum prices could use the model to forecast potential cost impacts and adjust procurement strategies or negotiate fixed-price contracts to mitigate risk. Similarly, a builder facing volatility in steel prices might leverage the model to decide whether to accelerate or delay structural work. Importantly, the model’s identified 2–4 quarter lag between petroleum price movements and BCI changes provides firms with a clear window to hedge fuel-intensive activities or renegotiate logistics and transport contracts before cost pressures materialise.
By incorporating economic indicators such as GDP and petroleum prices, the model allows professionals to make appropriate decisions, particularly valuable during economic volatility. Internationally, similar forecasting approaches have been applied to construction projects in the USA (Jiang et al., 2022) and New Zealand (Zhao et al., 2020), but these models often focus on input-based indices or specific projects such as highways. In contrast, the proposed output-based BCI model provides a more comprehensive view of total project costs, including profit margins and overhead, making it particularly relevant for building projects in Australia and adaptable for use in other regions.
For policymakers, the model provides a framework to assess how macroeconomic trends affect building construction costs, supporting decisions that strengthen the sector’s resilience. For example, knowing that crude oil prices significantly impact construction costs may prompt policy initiatives aimed at reducing energy costs or encouraging alternative energy sources in construction. In addition, the BCI model equips construction firms with a strategic tool to align pricing with market conditions, enhancing competitive advantage. By forecasting costs based on leading indicators, firms can anticipate how broader economic shifts may influence project costs, enabling them to strategically price their services. This foresight enhances their capacity to adjust pricing for profitability while staying competitive, ultimately leading to better project cost control and pricing adjustments that reflect true economic conditions.
6. Conclusion
While most studies on CCIs have focused primarily on general CCIs, they typically use input-based models, concentrating on costs of labour and other input resources. However, there remains a lack of research analysing output-based cost indexes that are tailored specifically to building projects. This study addressed this gap by identifying the leading indicators for the Building Cost Index and developing a model to forecast BCI, reflecting the unique economic pressures of the building sector. The findings revealed that key macroeconomic indicators – GDP, petroleum prices and PPIs of steel, electrical and other metals such as insulation and plaster products – serve as the most influential predictors of BCI. Statistical testing, including Granger causality and cointegration tests, confirmed that the VECM is suitable for forecasting BCI, and a VECM was subsequently developed to predict BCI based on these identified indicators.
This study contributes both theoretically and practically to the field of construction economics. Theoretically, it extends the application of producer theory within the building construction sector by illustrating how builders adjust pricing strategies in response to macroeconomic shifts and material-specific cost pressures. The research highlights unique indicators relevant to building projects, setting them apart from other construction types. In this way, it emphasises that pricing within the building construction sector is sensitive to broader economic conditions, and particularly to price shifts in specific construction materials, as captured by PPIs.
Practically, the forecasting model developed in this study holds substantial value for stakeholders involved in building projects. For construction managers, developers, policymakers and other decision-makers, the model offers a tool for anticipating future cost trends. By integrating macroeconomic factors into building cost forecasting, the model supports more accurate budgeting, enhances risk management and aids in planning resource allocation. Policymakers can leverage this model to assess the economic resilience of the construction sector, monitor price stability and make informed decisions about regulatory strategies. This focus on macroeconomic and material cost indicators allows industry professionals to make proactive decisions, potentially mitigating cost volatility associated with fluctuating economic conditions.
Nevertheless, while this study provides a novel approach to BCI forecasting, some limitations warrant future exploration. The model was developed using national-level data from Australia, which may limit its direct applicability in other countries or regions with distinct economic conditions, regulatory frameworks or construction practices. This limitation should be noted, and future research should replicate and adapt the model in other geographic and economic contexts to validate its robustness and enhance its broader applicability. In addition, future studies may consider developing similar models for industrial and infrastructure projects in Australia and other contexts. Beyond replication, future research could explore hybrid approaches that integrate VECM with machine learning techniques to capture nonlinear dynamics and improve predictive accuracy under structural breaks such as global crises. Importantly, this research signals a paradigm shift from generic, input-based cost forecasting towards project-specific, macro-integrated models that bridge theory and practice, enabling more accurate and context-sensitive predictions with significant economic and commercial impact.
