Applying Accepted Economic Indicators to Predict Cost Escalation for Construction

Introduction

The causes of construction cost escalation are many and complex (Yeo, 1990). They include labor equipment and material inflation, construction demand and market conditions, taxes and other government actions and major events. The ability of contractors to forecast reasonable estimates for construction is often considered the life-blood of the organization. The estimate is generally compiled by assembling fairly comprehensive quantity surveys and quotes for vendor and supplier items. Overhead and profit are then added based on the individual company’s requirements. The last item, contingency, is finally added to cover the cost of unknowns. These unknowns can be due to a variety of issues such as poor plans and factors affecting resources such as price escalations of materials and skilled labor over the life of the project.

 One way a contractor attempts to mitigate these cost increases is by use of a cost multiplier. This cost adjustment is often based on little deterministic factors and is often developed through experience. Estimating formula and mathematical models have attempted to modify this practice. Traditional business forecasting models for predicting cost increases are being considered and applied to construction estimating. These business models involve extrapolation of past data into the future by using linear and nonlinear curves or mathematical relationships (Ostwald, 2000). Predicting these escalations and identifying the variables that effect the increases can be a complex issue and is generally based on experience and judgment. There have been a number of attempts to predict these increases by utilizing linear regression, time-series, and neural network models. Each of these has some value but can be limited in their outcome.

The ability to successfully predict the cost escalation of labor and materials most widely used in construction is also required by a number of other industries. These industries often utilize economic indicators to develop business strategies and goals.  Economic indicators are also used to assist in determining the price of their products.  Users of the same materials utilized in construction namely; auto manufactures for steel and paper mills for lumber, apply prediction models for forecasting increases in material prices. Some of the variables utilized in econometrics modeling include various economic indicators. However, there are three composite indexes that are generally accepted for use in analyzing the trend of economic direction. The composite economic indicator models, specifically the leading, coincident, and lagging indexes will be analyzed to determine if there are any correlations between these and construction pricing over a period of time.

The identification of correlations between the economic indicators and costs of construction could benefit the contractors by providing an additional factor, a predictable trend, which may assist in forecasting cost escalations.

Defining the Variables

The most difficult task in predicting the cost increase of construction is the tremendous amount of variables that can impact the decision model. Research conducted comparing an artificial neural network model to the linear regression model and exponential smoothing model to predict construction costs, indicated that factors that influence construction prices are many and complex (Sinha & McKim, 1997). Because of this it can be difficult to reliably develop a prediction model without also accounting for fluctuations in the general economy.

In order to explore a potential correlation between indexes and construction pricing reliable data must first be collected for the cost of construction over time. There are a number of sources available that routinely collect and publish the cost of construction. Most notably are ENR’s Construction Cost & Building Cost indexes. Table 1 shows a sample of how the materials are tracked on a monthly basis. ENR has published cement, lumber, and steel prices since the early 1900’s. The indexes using these components are 1913=100 base year.

Table 1

Sample of ENR’s material tracking

      * Cost Indexes: Listed in ENR's Sep 23, 2002 Issue

The material cost indexes and labor cost indexes are then combined to develop a Construction Cost Index and Building Cost Index. These indexes include a great deal of cost data and will be the basis of comparison to the leading economic indicators. Predictability of cost increase and/or decrease may be possible by examining the data with statistical modeling to determine if there are correlations and trends that exists which can be used for forecasting cost escalation.

Construction Cost Prediction Models

In the past a number of statistical prediction models have been attempted for use in forecasting construction costs. Most of these models have proven unreliable and/or unusable by construction companies. This is due in part by a number of unknown variables and non-deterministic fluctuations that impact the traditional trend models.

A descriptive statistical model based on a single variable and its frequency distribution can aptly identify predictability. However, it does not account for the multiple variables and uncertainties that effect the forecasting of cost increases.  According to Sinha and McKim, regression models do consider multiple variables but have not proven to be useful tools for the construction contractor. These models fall short because they do not account for the relationship between independent and dependent variables and are nonlinear with unknown degrees. Artificial neural networks with multi-layer structures have also been explored as a method of prediction, but have also proven unsuccessful due to variables and their affects on prices.

In an attempt to develop a useful method for determining cost escalation, an alternative method will examine the economic composite indexes utilized by business forecasters and economists to determine if there is a correlation between these indexes and general construction costs. A study of the trends in index variation and correlation between these and the leading economic indicators could assist in the development of a useful prediction model.

Data Presentation

General construction costs (Construction Cost Index and Building Cost Index from ENR) and economic indicators (LEI, CEI and LEI from The Conference Board) were gathered and are displayed in Appendix 1, for a period of twenty four years on a quarterly basis ranging from September 1978 to September 2002. This time frame was selected in order to reflect a myriad of economic upturns, downturns, political influences, recessions, and business cycles.

According to The Conference Board, a not-for-profit non-advocacy business organization that researches and produces reports on business planning and development, we have experienced three recognized recession periods within the selected twenty-four year period. The first covered approximately March 1980 – June 1980. The second incident spanned from June 1981 – September 1982. The most recent recession period occurring June 1990 – March 1991. There is some debate on the specific length and severity of these recessions, but for this paper it is assumed that these are the accepted periods of economic trough.

The collected data was then used in order to evaluate the presence of any correlation patterns between the economic composite indicators and the general construction costs.  The first step requires definition and identification of the Economic Composite Indexes used to evaluate trends in the general economy.

Economic Composite Indexes

The composite leading, coincident, and lagging indexes are the key elements in an analytic system designed to signal peaks and troughs in the business cycle. Because they are averages, they tend to smooth out a good part of the volatility of the individual series and thereby serve as handy summary measures of the business cycle. Historically, the cyclical turning points in the leading index have occurred before those in aggregate economic activity, while the cyclical turning points in the coincident index have occurred at about the same time as those in aggregate economic activity. The cyclical turning points in the lagging index generally have occurred after those in aggregate economic activity. A change in direction in a composite index does not signal a cyclical turning point unless the movement is of significant size, duration, and scope. It is important to recognize that the timing of the leading index has been irregular and "false signals" are inevitable. The main value of the leading index is in signaling that either the risk of a recession has increased or that a recession may be coming to an end.

Although it is often stated in the press that three consecutive downward movements in the leading index signal a recession, economists do not endorse use of such a simple inflexible rule.  The January 1997 issue of Business Cycle Indicators discusses how a 1% decline (2% when annualized) in the leading index, coupled with declines in a majority of the 10 components, provides a reliable, but not perfect, recession signal.

The leading, coincident, and lagging (LCLg) indexes are an analytical system of assessing current and future economic trends, particularly cyclical expansions and recessions.  The system is based on grouping some key indicators according to their tendency to change direction before, during, or after the general economy turns from a recession to an expansion or from an expansion to a recession. In substance the leading index reflects business commitments and expectations, the coincident index reflects the current pace of the economy, and lagging index reflects business costs. The three indexes are called composite indexes because they group several component indicators (Frumkin, 1994).

The leading index components reflect the degree of tightness in the labor market due to employer hiring and firing; the buildup of orders and contracts that effect future production; materials prices that reflect shortages or gluts of raw materials used to expand or reduce existing inventories; financial conditions associated with the availability of funds in credit markets; and consumer psychology that effects household spending. This index can also be thought of as a gauge for the future state of the economy. Since it is comprised of a number of factors such as new orders, new housing, and consumer expectations, it can be used to analyze the near trend in economic output. Table 2 contains the index of leading economic indicators, which is a composite of the following ten specific indicators and their standardization factors.

Table 2

 Leading Economic Indicator Composite

The coincident index components reflect employment, real incomes generated from production, output in cyclically sensitive mining and manufacturing industries, and real manufacturing and trade sales depicting the flow of goods. This index can be thought of as a reflection of the current state of the economy. The following Table 3 indicates the four specific indicators used to compute the coincident composite.

Table 3

Coincident Index Composite

The last composite index utilized by economists to determine the trend in business is the lagging index.  This index’s components reflect: the effect of the duration of unemployment on business costs of recruitment and training, the cost of maintaining inventories, labor cost per unit of output, the burden of paying back business loans, interest payments as cost of production and prices of consumer services. Lagging index can be used as a tool to analyze the future trend in economy. Since it is comprised of indicators that could have an effect on the economy at a future time. Table 4 displays the seven specific indicators and standardization factors in computing the lagging economic index.

Table 4

Lagging Index Composite

Note:  The component standardization factors are inversely related to the standard deviation of the month-to-month changes in each component. They are used to equalize the volatility of the contribution from each component and are "normalized" to sum to 1. When one or more components are missing, the other factors are adjusted proportionately to ensure that the total continues to sum to 1. The index standardization factors are used to make volatility of the percent changes comparable for the three indexes. (TCP, 2002)

A study of the composite indexes was prepared which analyzed the quarterly reporting of indexes from June 1978 to June 2002. Figure 1 indicates the comparison of the leading, coincident and lagging indexes.  The LCLg indexes are based on 1996 =100.

As can be seen from the graph, the indexes have indicated a steady rise in the past ten years.  One anomaly that must be addressed occurred in 1996, which indicated a shift in rate of the coincident index, which caused it to overtake the leading and lagging indexes.  It is thought that this may have been due to a significant swing in the unemployment rate.

Trends in Construction Pricing

 In an effort to explore a potential correlation in construction prices and the composite economic indexes, prices were studied over the same twenty-four year period on a quarterly basis.  The information was compiled by utilizing ENR’s Quarterly Cost Reports on Construction Cost Index (CCI) and Building Cost Index (BCI) history from June 1978 through June 2002.

 ENR publishes two indexes that reflect the cost of construction in the United States. The first, Construction Cost Index is built by combining 200 hours of common labor at the 20-city average of common labor rates, plus 25 cwt of standard structural steel shapes at the mill price prior to 1996 and the fabricated 20-city price from 1996, plus 1.128 tons of Portland cement at the 20-city price, plus 1,088 board-ft of 2 x 4 lumber at the 20-city price. The second indicator, Building Cost Index is determined by combining 66.38 hours of skilled labor at the 20-city average of bricklayers, carpenters and structural ironworkers rates, plus 25 cwt of standard structural steel shapes at the mill price prior to 1996 and the fabricated 20-city price from 1996, plus 1.128 tons of Portland cement at the 20-city price, plus 1,088 board-ft of 2 x 4 lumber at the 20-city price.

The cities that ENR maintains cost data on are: Atlanta, Baltimore, Birmingham, Boston, Chicago, Cincinnati, Cleveland, Dallas, Denver, Detroit, Kansas City, Los Angeles, Minneapolis, New Orleans, New York, Philadelphia, Pittsburgh, St. Louis, San Francisco, and Seattle. Additionally, ENR reporters collect spot prices for all material tracked. The reporters survey the same suppliers each month in the targeted cities.  Actual prices may vary depending on the competitiveness of the market and local discounting practices. This method allows for a quick indicator of price movement (ENR.com). Further definition of the indexes indicates that both apply to general construction costs. The CCI can be used where labor costs contribute a high proportion of total costs. The BCI is more applicable to structures. Also according to ENR, the major difference between the two indexes is in their labor component. The ENR indexes measure how much it cost to purchase this hypothetical package of goods and services compared to what is was in the base year (ENR.com, 2002).

Figure 1 is a graphical representation of quarterly compilation of both indexes reported by ENR from June 1978 to June 2002 as was presented in Table 2. As can be seen by the chart, the indexes have been on a steady rise. There are some apparent spiked rises and declines over the periods but essentially constant rates of increase have occurred. Linear trend lines were determined using the calculated least square fit represented by the equations. An R-squared value was also included in order to further define the fit of the trend line to the data. Both of these calculations were accomplished using the statistical formulas built into Microsoft Excel.

Correlation of Data

 In order to study the collected data to determine if there is a correlation between CCI, BCI and Composite Economic Indicators the equations for simple correlation, as applied by Microsoft Excel, for determination of correlation coefficient was used. This equation returns the correlation coefficient of the array1 (CCI or BCI)) and array2 (leading, coincident, and lagging indexes). Use of this correlation coefficient will determine the relationship between two properties. The following results were determined.

Table 5

Correlation Coefficient Comparisons of CCI & BCI

A 1.0 correlation coefficient indicates a perfect correlation between two data sets. As can be seen from Table 5 there are indications of high correlations. Specifically between the CCI, BCI and also the Leading and Coincident Indexes. However, it is the square of the coefficient of correlation that is the ratio of the explained variation of the total variation (Wheelwright, 1973). This value is referred to as the coefficient of determination.

Table 6

Coefficient of Determination Comparisons of CCI & BCI

Table 6 further indicates a high correlation between the CCI, BCI and Coincident Index. The CCI and Coincident Index coefficient of determination of (Correl)² = .9708 indicates that 97.08% of the 98 data samples that were used in fitting the regression line to these observations of the variation from mean value of Y, were explained by that regression line. The results indicate that there are no significant differences or variances in the correlation of data.

An additional study was performed that plotted the Construction Cost Index versus the Building Cost Index. The following Figure 2 illustrates the results.

As can be observed, the graph also appears to indicate a high degree of correlation between the two sets of data. The formula for the trend line and R-square analysis, close to 1.0 or perfect correlation, suggests that these indexes are constant in their relational value over time.

Since the previous comparisons of CCI, BCI and Leading indicated a relatively high correlation; the following relational graph, Figure 3, was prepared in order to analyze the relationship between the CCI and Leading Index. The study also indicates a high correlation and relational dependency over time. The trend line formula and R-squared analysis show a .9008 correlation, a fairly high value of dependency and relation.

In an effort to further qualify the dependency relations, a graphical study was conducted comparing the Lagging Index to the CCI. It was observed in Tables 5 & 6 that there was a much lower computed correlation between these than those of the CCI and Leading Index. Figure 4 shows that this is in fact true. In plotting the CCI versus the Lagging Index there is a discernable difference and appears scattered as opposed to linear in a trend analysis. The trend line and R-squared calculations indicate .2338, a relatively low correlation factor.

Finally one additional graphical study was performed that compared the CCI and the Coincident Index. As indicated previously, the Coincident Index should indicate the highest level of correlation based off the Table 4 &5 calculated values. This would also be consistent with the presumption that this index measures the current state of the economy and should be reflective and representative of the current state of CCI. As can been seen in Figure 5. This is indeed the case. The trend line and R-squared analysis does suggest a high level of correlation. R-squared relation in this set is .9863 close once again to the perfect correlation factor of 1.0.

Conclusions

The results of the correlation study and graphical comparison study do indicate a relationship between the cost indexes and the economic composite indexes. The data suggests that predictability of Construction Cost Index and Building Cost Index can be accomplished by monitoring the changes in the Leading Index. The Lagging Index also indicates some relationship but did not demonstrate a high correlation when graphically compared and analyzed. Not surprisingly the Coincident Index was determined to have the highest correlation even when observed graphically it did indicate a relationship to the CCI.

By monitoring predictions and trends of composite indexes by economists, the constructor should have a greater indication of the trends and rates of changing construction costs. This information could be useful in helping to estimate escalation of costs over the life of a project.

 It is realized that utilizing economic indexes does have limitations. First, the preliminary cotemporaneous data that are available during months before the downturn into a recession or upturn into recovery do not always provide advanced signs of a cyclical turning point. Second, the LCLg indexes do not forecast quantitative economic growth rates or time of future cyclical turning points. Third, the indexes occasionally give false signals of a pending change in the direction of the economy (Frumkin, 1994). Additionally, David Orr, Chief economist at First Union Corp. in Charlotte, N.C. (Daily Reporter, 2001) noted that the market tends to discount the report on leading indicators because performance is widely known, by economists, before the index itself is calculated. This, he said, “misses its influence on the perceptions of the general public and politicians”.

In light of these limitations it still can be deduced that there is a correlation between LCLg and long-term general direction of costs for construction. Constructors must generally guarantee a price with little factual knowledge of the changing economic conditions and how it could affect their prices throughout the cost of a project. With the understanding that published and accepted economic indicators are closely linked to construction cost indexes, better pricing decisions can be realized. Knowledge of this can assist in the decision and cost prediction models by contributing a variable that can be utilized when predicting escalation of costs.

One of the perceived values of applying prediction models to forecast costs is in its use as a marketing tool. A construction company that can base its estimates of future construction on statistical analysis, as opposed to those that base it solely on prescribed escalation factors, will provide a “value added service” to their clientele. The clients will realize a higher degree of credibility and reputability when analyzing escalating costs for projects, which in turn should yield higher sales for the contractor.

References

Hanna, A. S. and Chao, L., (1994). “Quantification of Cost Uncertainties Using Neural Network Technique,” Pro., 1^st Congress on Computing in Civ. Engrg., Washington D.C. Vol. I, p41-46.

Frumkin, N. (1994). Guide To Economic Indicators, second edition, M.E. Sharpe.

ENR.com, “ENR Cost Indexes FAQ’s” http://www.enr.com/cost/costfaq.asp 8/9/00

Ostwald, Phillip F.(2000). Construction Cost Analysis and Estimating, Prentice Hall

Powell, Eileen, “Revised Index of Leading Indicators Drops 0.6 Percent in December”, The Daley Reporter, Tuesday January 23, 2001, Vol 13, Number 16

Sinha, Sunil K.; McKim, Robert A. (1997). “Forecasting Construction Cost Escalation Using Artificial Neural Networks,” Pro., 1997 Artificial Neural Networks in Engineering Conference, ANNIE’97, St. Louis, MO. Vol. 7, p829-835

TCB Indicators, “Leading Economic Indicators & Related Composite Indexes”

http://www.tcb-indicators.org/, 6/1/2002

The Conference Board Inc., “Business Cycle Indicators”, http://www.globalindicators.com/, 1997

 Wheelwright, S. and Makridakis, S., Forecasting Methods for Management, John Wiley & Sons, 1973

Yeo, K.T. (1990). “Risks, Classification of Estimates, and Contingency Management,” Journal of Management in Engineering, ASCE 10(1), p 72-84

Appendix

General construction costs (Construction Cost Index and Building Cost Index from ENR) and economic indicators (LEI, CEI and LEI from The Conference Board) were gathered and are displayed in Appendix 1 for a period of twenty-four years on a quarterly basis ranging from June 1978 to September 2002.

Note: Highlighted areas are indication of recession years

Appendix (cont.)