**Suggested Citation:**"5. Cost Forecasting Approaches." National Academies of Sciences, Engineering, and Medicine. 2020.

*Improving Mid-Term, Intermediate, and Long-Range Cost Forecasting for State Transportation Agencies*. Washington, DC: The National Academies Press. doi: 10.17226/25974.

**Suggested Citation:**"5. Cost Forecasting Approaches." National Academies of Sciences, Engineering, and Medicine. 2020.

*Improving Mid-Term, Intermediate, and Long-Range Cost Forecasting for State Transportation Agencies*. Washington, DC: The National Academies Press. doi: 10.17226/25974.

**Suggested Citation:**"5. Cost Forecasting Approaches." National Academies of Sciences, Engineering, and Medicine. 2020.

*Improving Mid-Term, Intermediate, and Long-Range Cost Forecasting for State Transportation Agencies*. Washington, DC: The National Academies Press. doi: 10.17226/25974.

**Suggested Citation:**"5. Cost Forecasting Approaches." National Academies of Sciences, Engineering, and Medicine. 2020.

*Improving Mid-Term, Intermediate, and Long-Range Cost Forecasting for State Transportation Agencies*. Washington, DC: The National Academies Press. doi: 10.17226/25974.

**Suggested Citation:**"5. Cost Forecasting Approaches." National Academies of Sciences, Engineering, and Medicine. 2020.

*Improving Mid-Term, Intermediate, and Long-Range Cost Forecasting for State Transportation Agencies*. Washington, DC: The National Academies Press. doi: 10.17226/25974.

**Suggested Citation:**"5. Cost Forecasting Approaches." National Academies of Sciences, Engineering, and Medicine. 2020.

*Improving Mid-Term, Intermediate, and Long-Range Cost Forecasting for State Transportation Agencies*. Washington, DC: The National Academies Press. doi: 10.17226/25974.

**Suggested Citation:**"5. Cost Forecasting Approaches." National Academies of Sciences, Engineering, and Medicine. 2020.

*Improving Mid-Term, Intermediate, and Long-Range Cost Forecasting for State Transportation Agencies*. Washington, DC: The National Academies Press. doi: 10.17226/25974.

**Suggested Citation:**"5. Cost Forecasting Approaches." National Academies of Sciences, Engineering, and Medicine. 2020.

*Improving Mid-Term, Intermediate, and Long-Range Cost Forecasting for State Transportation Agencies*. Washington, DC: The National Academies Press. doi: 10.17226/25974.

**Suggested Citation:**"5. Cost Forecasting Approaches." National Academies of Sciences, Engineering, and Medicine. 2020.

*Improving Mid-Term, Intermediate, and Long-Range Cost Forecasting for State Transportation Agencies*. Washington, DC: The National Academies Press. doi: 10.17226/25974.

**Suggested Citation:**"5. Cost Forecasting Approaches." National Academies of Sciences, Engineering, and Medicine. 2020.

*Improving Mid-Term, Intermediate, and Long-Range Cost Forecasting for State Transportation Agencies*. Washington, DC: The National Academies Press. doi: 10.17226/25974.

**Suggested Citation:**"5. Cost Forecasting Approaches." National Academies of Sciences, Engineering, and Medicine. 2020.

*Improving Mid-Term, Intermediate, and Long-Range Cost Forecasting for State Transportation Agencies*. Washington, DC: The National Academies Press. doi: 10.17226/25974.

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

52 5. Cost Forecasting Approaches 5.1 Introduction The protocol proposed in Chapter 4 for the comparative analysis of cost indexing alternatives is only intended to identify the index that most effectively represents the local construction market. After successfully applying the protocol to identify the most suitable cost indexing alternative, the agency still has to face the challenge of selecting an appropriate method to generate an inflation rate from that index. The research efforts presented in this chapter are aimed to assist STAs with that challenge. The chapter summarizes the process and findings associated with the assessment of various cost forecasting approaches. That assessment facilitated a better understanding of their forecasting performance and allowed the team to formulate guidelines to maximize their effectiveness. Recognizing that there is some variability in the level of effort and resources that different STAs would invest towards the improvement of their cost forecasting practices, this study considered different cost forecasting alternatives to provide guidance that accommodates the preferences and requirements of different agencies. Those alternatives range from the use of standard inflation rates with no quantitative analysis of the local construction market to an advanced data-driven procedure proposed by the research team called Moving Forecasting Error (MFE). This innovative method was first used in this study to evaluate other approaches, but its results were later used to generate inflation rates and risk-based forecasting time horizons. The study also evaluated the cost forecasting performance of two regression analysis approaches: linear and exponential regression. Although Chapter 4 demonstrated the superior performance of Multilevel Construction Cost Indexes (MCCIs) over traditional in-house and external cost indexing alternatives, traditional approaches could still be the preferred option for some STAs. Therefore, the assessment of the forecasting approaches mentioned in the previous paragraph was conducted using various cost indexing alternatives. Cost forecasting approaches were evaluated on their forecasting effectiveness over mid-term, intermediate-range, and long-term forecasting time horizons. Their performance was evaluated for each case studies agency on two different types of work: asphalt and concrete paving. Further research is needed to replicate this study for other types of work. 5.2 Standard Annual Inflation Rates The assessment of cost forecasting approaches started with the simplest approach; the use of standard annual inflation rates, which seems to be the preferred option for most STAs. Those rates are not the result of an in-house assessment of the local construction market. They are usually adopted following recommendations from the Federal Highway Administration (FHWA), other federal or state agencies, or financial consultants. Those rates are used in all projects and programs regardless of their specific requirements and anticipated scopes of work and ignoring their associated underlying implications. Common standard inflation rates currently used by STAs were found to be between 3% or 4%. The study also found examples of these rates being applied as

53 simple or compounded rates. Section 1.4.2 in Chapter 1 explains the difference between simple and compounded inflation rates. The research question formulated by the research team to assess the performance of standard inflation rates was: what would have been the performance of different standard inflation rates across the 20 years of available historical bid data? To answer this question, the research team used the proposed MFE methodology, which compares market projections given by different standard inflation rates against actual market fluctuations observed within the 20-year datasets. For each case study region, and in the absence of better indicators of actual construction market fluctuations, the study assumed that the best MCCI identified for each region in Chapter 4 provides a fair representation of its respective local construction market. Therefore, the effectiveness of standard inflation rates, as well as the effectiveness of the other cost forecasting approaches, was measured in terms of how good they were at predicting index values along those MCCIs. It should be noted that if a better indicator were found for a given region, this would become the most suitable index for that region instead of the MCCI identified in Chapter 4. The MFE methodology is an iterative process designed to maximize the value of the available data. Unfortunately, it is difficult to find STAs with more than 20 years of available historical bid data in a suitable format for data processing and analysis. Thus, the study had to rely on single 20- year datasets to evaluate the performance of different forecasting approaches. Some traditional data analysis methods used in cost estimating research would probably generalize any findings resulting from this single 20-year dataset, but the MFE method recognizes that there are a number of 3, 5, 10, and 15-year periods within the available data and takes advantage of those smaller data partitions. For instance, the first index value of the MCCIs developed for the Delaware Department of Transportation (DelDOT) was set in July 1999. If the first index value from one of DelDOTâs MCCI cost indexes were projected three years into the future with a 3% annual inflation rate, the comparison of the projected value and the actual index value on July 2002 would provide an idea of the cost forecasting error that DelDOT could expect from the use of a 3% inflation rate over a three-year forecasting period. If this three-year window of time were then moved six months forward, starting at the next indexing period (January 2000), and the error for a three-year forecast were calculated again with the same inflation rate, it would provide a second measure of error for the same forecasting scenario. This process could be repeated 34 times over a 20-year period, generating 34 cost forecasting errors for a 3% annual inflation rate over a three-year forecasting period. This would offer a better understanding of the risk associated with that forecasting scenario. Similarly, 30, 20, and 10 iterations could be performed with 5-, 10-, and 15-year time windows, respectively. Since the first index value is calculated after the first six months of data, the available dataset would not even allow the calculation of a single forecasting error for a 20- year forecast. The longest possible forecast that could be assessed would be over 19.5 years, and only one forecasting error could be generated for that scenario. Instead of relying on this single long-range forecasting error, the proposed MFE method takes advantage of the more reliable assessments conducted for shorter forecasting periods, identifies trends, and extrapolate those trends to the long-range forecasting zone. Figure 5.1 illustrates the

54 capabilities of the MFE methodology. The figure shows the performance of a 4% compounded annual inflation rate on an asphalt paving scope of work in Colorado. Each point in this figure corresponds to the average forecasting error calculated for each forecasting period from 0.5 to 19.5 years. Figure 5.1 Example of MFE Output â Average Forecasting Errors for DelDOTâs Asphalt Paving Activities with a 4% Compounded Annual Inflation Rate. As explained above, the number of error measurements used to calculate each average value in Figure 5.1 decreases as the forecasted time horizon increases. The average error for 0.5 years was calculated with 39 values, while the one for 19.5 years is not actually an average. That is the only measured error for that period of time. The figure shows how the iterative MFE process was able to model a trend for the change in average percentage errors over time. The negative sign in the average forecasting errors means that actual market values tend to be lower than those obtained with the inflation rate under consideration. Figure 5.1 corresponds to an asphalt paving scope-based CCI developed with the most suitable MCCI identified for the Northeast Region in Colorado. Similar linear trends were found in all case study regions for both asphalt and concrete paving. In most cases, points tend to start falling off the trend after about 15 years, where average values start to be calculated with less than 10 observed forecasting errors. Thus, the MFE method was designed to ignore all values calculated with less than 10 observations, and to use regression analysis to project the remaining values into the future to estimate expected errors for long-range forecasts. That is how the linear projection in Figure 5.1 was created. The MFE method not only facilitated a better assessment of average forecasting errors, but also allowed the projection of percentiles around average values to establish error ranges at 50%, 70%, and 90% confidence levels. Figure 5.2 shows the same linear trend of average errors from Figure 5.1, but this time with its respective confidence intervals. Based on this figure, the Colorado Department of Transportation (CDOT) could reasonably assume, with a 90% confidence level, that any 15-year asphalt paving cost forecast estimated in the Northeast Region with a 4% compounded inflation rate would offer a forecasting error between +12% and -27%. It should be

55 noted that forecasting errors in this study do not include uncertainties associated with current- dollar estimates. Average errors in Figure 5.1 are only associated with forecasting risks. The level of error for unforecast estimates should be assessed through a separate process. The Cost Forecasting Toolkit developed by this study assists with the integration of pre- and post-forecasting estimating errors to generate more effective risk-based forecasting timelines. Figure 5.2 Example of MFE Output â Average Forecasting Errors with Confidence Intervals for DelDOTâs Asphalt Paving Activities with a 4% Compounded Annual Inflation Rate After assessing the performance of a range of possible standard inflation rates for asphalt and concrete paving across all case studies and regions, and assuming that the three case study agencies could reasonably represent all STAs, the research team proceeded to synthesize the results of those assessments into a set of standard inflation rates. Those standard rates are consolidated into Table 5.1. They are intended to be used by agencies that decide not to conduct an internal analysis of the local construction market to generate suitable inflation rates. Those agencies would be relying on other STAsâ experiences. Table 5.1 corresponds to Module 2 of the Cost Forecasting Approach Selection Framework presented in the Transportation Cost Forecasting Guidebook. Table 5.1 Consolidation of Standard Inflation Rates from Case Studies Low Medium High Low Medium High Low Medium High 2% 3%â4% 5% 2% 3%â4% 5% 3% 4% 5% @50%Â CL @70%Â CL @90%Â CL 2% 3% 4% 2% 3% 4% 2% 3% 4% @50%Â CL @70%Â CL @90%Â CL RecommendedÂ Rates ExpectedÂ ForecastingÂ Errors ASPHALTÂ PAVING Â±15% Â±15% â20%Â toÂ 30% Â±20% Â±20% â25%Â toÂ 35% Â±30% Â±30% â30%Â toÂ 45% Â±15% RecommendedÂ Rates CONCRETEÂ PAVING ExpectedÂ ForecastingÂ Errors ForecastingÂ TimeÂ Horizon MidâTermÂ (3â5Â years) IntermediateâRangeÂ (10â15Â years) LongâRangeÂ (20Â Years) TypeÂ ofÂ InflationÂ Rate AnnualÂ Simple AnnualÂ Compounded AnnualÂ Compounded InflationÂ ExpectedÂ Level Â±35% â30%Â toÂ 35% â35%Â toÂ 40% â40%Â toÂ 50% â25%Â toÂ 30% â30%Â toÂ 40% â40%Â toÂ 45% Â±25%

56 Results from the case studies not only showed significant differences in inflationary trends among geographic regions within the state, but also between agencies. The three case study agencies were found to represent inflation trends at three different levels of magnitude: low, medium, and high. Table 5.1 provides standard low, medium, and high annual inflation rates for each of the three forecasting time horizons under consideration (mid-term, intermediate, and long-range). This facilitates more effective guidance for STAs on the low and high ends of the spectrum. Table 5.1 also provides forecasting error ranges at three different confidence levels for the three forecasting time horizons and for both scopes of work. The use of medium annual inflation rates could be used by agencies that do not have reliable information to infer the level of magnitude of upcoming inflation rates. However, it should be noted that the forecasting error ranges in Table 5.1 are applicable under the assumption that the agency would appropriately place itself in one of the inflation magnitude categories. An error in doing so would increase the level of uncertainty, widening the forecasting inflation rate. Although Table 5.1 constitutes an improvement in the quality of guidance for users of standard inflation rates, a better cost forecasting performance can still be achieved by calculating annual inflation rates through more formal quantitative procedures using the guidelines provided in the rest of this chapter and in NCHRP Research Report 953. Those guidelines allow an effective calculation of annual inflation rates on a case-by-case basis, resulting in a considerable reduction of cost forecasting uncertainty. 5.3 Risk-Based Forecasting Timelines from MFE Results The research team found that projections of MFE cost forecasting errors, like those shown in Figure 5.2, can easily be transformed into risk-based forecasting timelines (see Section 1.3) for any current-dollar estimate by following some simple steps described in NCHRP Research Report 953. Figure 5.3 shows the risk-based forecasting timeline created for a hypothetical asphalt paving program with a current-dollar estimate of $10 million. This risk-based output was generated by applying the cost forecasting errors from Figure 5.2 to the forecasted values that would be obtained with a 4% compounded projection. Figure 5.3 also shows the forecasted values that would be obtained with that compounded inflation rate. A more detailed description of the process to produce the risk-based forecasting timeline in Figure 5.3 with the forecasting errors from Figure 5.2 is presented in NCHRP Research Report 953.

57 Figure 5.3 Example of Risk-Based Forecasting Timeline with 4% Compounded Projection It seems evident from Figure 5.3 that a 4% compounded inflation rate is too high for the asphalt paving market in the Northeast Region of Colorado. The projection with that inflation rate increasingly deviates from the average forecast as the time horizon increases. Assuming that the target inflation rate is intended to match the average forecast in Figure 5.3, the next step would be to find the rate that yields a projection that most closely overlaps the average forecast in the risk- based forecasting timeline. The shape of the risk-based output for this example suggests that the average trend would be better matched by a compounded annual inflation rate. Figure 5.4 shows that the average forecast would be matched by a 3.1% compounded annual inflation rate, which completely overlaps the average expected trend. This would be the inflation rate suggested by the MFE method for the estimate used in this example. $5,000,000 $10,000,000 $15,000,000 $20,000,000 $25,000,000 $30,000,000 $35,000,000 0 5 10 15 20 25 30 Fo re ca st ed Â C os tÂ E st im at eÂ ($ ) ForecastingÂ TimeÂ HorizonÂ (Years) 90%Â ConfidenceÂ Level 70%Â ConfidenceÂ Level 50%Â ConfidenceÂ Level AverageÂ Forecast 4%Â CompoundedÂ Projection

58 Figure 5.4 Example of Risk-Based Forecasting Timeline with 3.1% Compounded Projection Risk-based forecasting timelines were developed and analyzed for both asphalt and concrete paving in all regions of the three case study agencies. The MFE methodology was applied using the most suitable MCCI, as well as the most suitable existing traditional CCI identified for each region. In all cases, the long-range behavior of the construction market showed an exponential behavior, suggesting that compounded inflation rates are always more appropriate than simple rates. However, as explained in Section 1.4.2, there is no significant difference in applying a simple and compounded inflation rate for mid-term forecasts. Therefore, the Cost Forecasting Approach Selection Framework resulting from this study considers the use of simple inflation rates for mid- term forecasts since a linear regression process would be easier to understand and model by most estimators. Case study results for all regions, all forecasting time horizons, and all cost indexing approaches were analyzed to produce the forecasting error ranges shown in Table 5.2. These ranges represent the consolidated performance of the MFE methodology across the three case studies. A comparison between MCCI forecasting error ranges in Table 5.2 and those shown before in Table 5.1 shows the considerable reduction in cost forecasting uncertainty that an STA could achieve if the proposed MCCI and MFE methodologies are implemented. $5,000,000 $10,000,000 $15,000,000 $20,000,000 $25,000,000 $30,000,000 $35,000,000 0 5 10 15 20 25 30 Fo re ca st ed Â C os tÂ E st im at eÂ ($ ) ForecastingÂ TimeÂ HorizonÂ (Years) 90%Â ConfidenceÂ Level 70%Â ConfidenceÂ Level 50%Â ConfidenceÂ Level 3.1%Â CompoundedÂ Projection

59 Table 5.2 Consolidated Forecasting Error Ranges from the Application of the MFE Method MCCI forecasting error ranges in Table 5.2 actually correspond to worst-case scenarios found among case study results on both ends of each range. An STA would obtain narrower forecasting error ranges after the actual application of the MFE method with its own data. Results obtained from the application of traditional CCIs are not as promising as those provided by the MCCI. However, those also represent the worst scenarios across all case study regions. A number of regions showed narrower error ranges with traditional CCIs than those shown in Table 5.1. A more reliable forecasting error range would also be obtained after using the MFE on the intended CCI. The assessment of cost forecasting procedures presented in this chapter was conducted with 20 years of historical bid data from each case study agency. Therefore, an STA implementing the cost methodologies proposed in this report would be expected to obtain results within the forecasting error ranges shown in Table 5.2, only if those methodologies are used with at least 20 years of historical bid data. In order to determine the implications of using a smaller dataset, the research team repeated the case study with the Minnesota Department of Transportation (MnDOT), assuming that only the most recent 10 years of data were available. This reduced case study showed an increase of about 40% in the forecasting error ranges shown in Table 5.2. Risk-based forecasting timelines should not be confused with results from regression analyses. A regression model is obtained from a single pass along the time series, while the proposed risk- based outputs are the result of a most exhaustive analysis that considered all possible time horizons contained within the available data. In other words, the 10 first years of the risk-based forecasting timeline are the result of analyzing all possible 10-year periods in the 20-year dataset, including bid data from projects awarded during the most recent 10 years. 5.4 Linear and Exponential Regression Analysis Linear and exponential regression analyses are mathematical methods commonly used to assess simple and compounded inflation trends, respectively. Their main difference lies in the assumed relationship between the time variable and the cost index selected to represent the construction market (a linear or an exponential relationship). Figures 5.5 and 5.6 show the simple and compounded annual inflation rates obtained from a linear and an exponential regression model developed for the same asphalt paving scope-based CCI. This CCI was developed with the most suitable MCCI for the northern regions of Minnesota. Statistical software packages do not usually directly provide simple and compounded inflation rates from linear and exponential regression MCCI TraditionalÂ CCI MCCI TraditionalÂ CCI MCCI TraditionalÂ CCI @50%Â CL Â±12% â20%Â toÂ 15% â10%Â toÂ 15% â30%Â toÂ 15% Â±10% â40%Â toÂ 10% @70%Â CL Â±20% â25%Â toÂ 20% â15%Â toÂ 20% â35%Â toÂ 20% Â±15% â45%Â toÂ 15% @90%Â CL Â±30% â35%Â toÂ 30% â25%Â toÂ 30% â45%Â toÂ 30% Â±25% â50%Â toÂ 20% @50%Â CL Â±10% â15%Â toÂ 25% Â±10% â15%Â toÂ 50% Â±10% â15%Â toÂ 55% @70%Â CL Â±15% â20%Â toÂ 23% Â±15% â20%Â toÂ 55% Â±15% â20%Â toÂ 60% @90%Â CL Â±25% â30%Â toÂ 40% Â±25% â30%Â toÂ 65% Â±25% â25%Â toÂ 75% CONCRETEÂ PAVING ExpectedÂ ForecastingÂ Errors ExpectedÂ ForecastingÂ Errors TypeÂ ofÂ InflationÂ Rate AnnualÂ Simple AnnualÂ Compounded AnnualÂ Compounded TypeÂ ofÂ CostÂ Index ASPHALTÂ PAVING ForecastingÂ TimeÂ Horizon MidâTermÂ (3â5Â years) IntermediateâRangeÂ (10â15Â years) LongâRangeÂ (20Â Years)

60 analyses. NCHRP Research Report 953 provides guidance to calculate inflation rates from the typical outputs provided by those software packages. Figure 5.5 MnDOT Asphalt Paving CCI â North Region â Linear Regression Model Figure 5.6 MnDOT Asphalt Paving CCI â North Region â Exponential Regression Model A comparison of the R-Squared values of the regression models in Figures 5.5 and 5.6 show that an exponential equation would better fit this index. As with the other parts of the quantitative analysis conducted for this study, these regression models were developed and analyzed for asphalt and concrete paving scope-based CCIs in all case study regions. As occurred with the MFE, it was clear that compounded inflation rates are a better fit for long-range time frames. The main difference between the proposed MFE and regression analysis techniques lies in their assumptions of risk. MFE can be classified as a more conservative or risk-averse approach since it produces a cost forecasting output that combines results from several forecasting scenarios created within the available data. On the other hand, regression models are the result of a single 80 130 180 230 280 330 6/29/1999 3/25/2002 12/19/2004 9/15/2007 6/11/2010 3/7/2013 12/2/2015 8/28/2018 In de xÂ V al ue Date Simple Annual Inflation Rate = 10.21% R2 = 0.863 80 130 180 230 280 330 6/29/1999 3/25/2002 12/19/2004 9/15/2007 6/11/2010 3/7/2013 12/2/2015 8/28/2018 In de xÂ V al ue Date Compounded Annual Inflation Rate = 5.4% R2 = 0.877

61 configuration of the available data, making them more appropriate for risk-seekers that decide to rely on a single scenario, underestimating cost forecasting uncertainties. However, the study also found that regression analysis techniques are more suitable to model the anticipated continuation of short-term abnormal price fluctuations. For example, regression analysis should be the preferred option for a mid-term forecast if the selected cost index measures an abnormal deflation situation during the last three years, and if there are strong reasons to believe that this trend will continue during the next three to five years. If that were the case, the look-back period for the development of the regression model should only cover the period with the abnormal pricing behavior. An MFE model in that case could yield a more conservative output opting for a more normal pricing behavior, ignoring the deflation behavior. Those types of abnormal market conditions are commonly referred to as market corrections. The study found that appropriate use of regression analysis techniques to model market corrections could improve the performance of the forecasting error ranges in Table 5.2 by 60% and 40% for mid-term and intermediate-range forecasts, respectively. When using regression analysis to project market corrections into the future, it is important to consider that most corrections last less than five years (according to observations form this study). Thus, the projection of an observed three-year downward trend for more than two years into the future would anticipate an unlike scenario. Likewise, the usual length of abnormal market conditions makes it inappropriate to forecast construction costs with five years of historical bid data, or less. A five-year look-back period could contain a downward trend that could be projected into the future over a mid-term, intermediate, or long-range forecasting period, representing also unlikely scenarios. To avoid this issue, this report suggests the use of at least 10 years of historical cost data for mid-term and intermediate-range forecasts, and, ideally, 20 years look-back periods for long-range forecasting procedures. 5.5 Chapter Findings Some of the most relevant results and findings from the research activities described in this chapter are synthesized in Tables 5.1 and 5.2. The following are some additional findings that resulted from the analysis of the results in those tables and some other relevant observations made during the assessment of cost forecasting alternatives. ï· STAsâ inflationary conditions can be classified as low, medium, and high. Table 5.1 suggests some standard annual inflation rates that could be used for each of those inflationary conditions. Correct classification of STAs among those three categories would considerably increase the effectiveness of cost forecasting processes for agencies that prefer the use of standard inflation rates without a quantitative assessment of the local construction market. ï· Transportation construction prices in all case study regions, and for both asphalt and concrete paving, have been increasing on an overall exponential rate during the last 20 years. This means that compounded inflation rates tend to be more suitable for cost forecasting purposes. However, significant differences between simple or compounded

62 inflation rates start to appear as the forecasting time horizon extends over more than 5 years. There are no significant differences in the use of simple or compounded inflation rates for mid-term forecasts. Therefore, and given that the linear regression process is slightly more straightforward than the exponential regression approach, the Cost Forecasting Approach Selection Framework gives preference to the use of simple inflation rates for mid-term forecasts and compounded rates for intermediate and long-range time frames. ï· The main difference between the use of an externally recommended standard inflation rate and one internally calculated from the analysis of historical market data is the level of uncertainty associated with the resulting forecasted cost estimate. The implementation of the proposed MFE, as well as other practices discussed in this chapter, could significantly reduce uncertainty levels in cost forecasting procedures through the generation of effective inflation rates. ï· The 10-year case study conducted in Minnesota showed that an agency with only 10 years of available historical bid data to implement the data-driven forecasting procedures proposed in this report could expect a 40% reduction in forecasting effectiveness with respect to the performance offered by a 20-year dataset. However, this statement was the result of a single reduced case study. Future research efforts could be aimed to validate this statement by replicating this analysis among other STAs. ï· The study found that the overall growth of construction prices follows an upward trend over the intermediate and long-range time horizons with most short-term deflation situations or market corrections lasting less than 5 years. ï· Cost forecasting processes with short look-back periods are more susceptible to market corrections, with the risk of anticipating unlikely future market scenarios. To minimize this risk, the study suggests the use of at least 10 years of historical cost data for mid-term and intermediate-range forecasts, and, ideally, 20 years look-back periods in long-range forecasting procedures. ï· In comparison with regression analysis, MFE could be described as a more conservative or risk-averse approach since its output is the result of the assessment of several forecasting scenarios created within the available data, while regression analysis relies on a single pass along the available data. However, the study found that regression techniques are more suitable to model the anticipated continuation of short-term market corrections. In those cases, MFE calculations could yield a more conservative output assuming a more normal pricing behavior.