Forecasting Models

1
Forecasting Models

• Forecasting Change in the Construction Industry
• By David Walls

2
Austin Commercial

• Large Construction Manager
• Based in Texas
• Operations throughout the United States
• Past customers include
• Intel
• Texas Instruments
• Exxon
• EDS
• FED
• SMU

3
Austin Commercials Problem

• Problem Dealing with change in construction
• Large amount of changes taking place
• Not taking into account the impact the changes
  have later in a project
• Uncertainty in knowing what was causing change
  and what impacts change had

4
Problem Analysis

• Discussions with Austins Management
• Two main indicators of change in a construction
  project
• RFIs (Request for Information)
• New drawings issued
• Austins needs
• A way to predict potential changes at any given
  point in project
• Impact of those changes on cost

5
Research and Data Collection

• Designed a spreadsheet to collect data
• RFIs and new drawings broken down monthly over
  the life of the project
• Broken down into Divisions
• Architectural
• Structural
• Civil
• Mechanical
• Electrical
• Cost Information
• Compiled a list of Project that were wanted

6
Spreadsheet
7
Projects

• Akin, Gump, Strauss, Hauer, and Field Project
• Alcon Laboratories Building G Project
• Austin Ventures Project
• CarrAmerica Project
• Clark, Thomas, and Winters Project
• Crossmark Project
• CTW Storage/Fitness Center Project
• Ft. Worth Convention Center Phase 1 Project
• Ft. Worth Convention Center Phase 2 Project
• Hall Office Project
• Love Field CUP Project
• Mabel Peters Caruth Center Project
• Terrace V Project (RFI info only)
• TriQuint Semiconductor Project
• University of North Texas Recreation Center
  Project
• University of Texas Southwestern Medical Center
  Project

8
Situation Analysis

• Calculated the Percentage of RFIs and new
  drawings that were complete at key points in a
  project (10 25 50 75 100)
• By division and Total
• Decided to use regression modeling
• Could obtain the most accurate fit of the
  relationship between the inputs and outputs of
  the project

9
Regression Models

• Two Regression Models
• Cubic (polynomial) Regression Model
• best fit line (cubic) for the relationship
  between the percentage complete in the job and
  the percentage of the RFIs or new drawings
  issued out of the total
• Multiple Regression Model
• Best fit (linear) for the relationships between
  the costs of a project and totals for RFIs and
  new drawings and initial budget

10
Cubic Regression Model

• Used Minitab to solve a cubic regression model
• For total RFIs and by division
• For total new drawings and by division
• Allow forecast of total RFIs and total new
  drawings by division at the end of the project

11
Example
12
RFIs Compared
13
New Drawing Compared
14
Multiple Regression Models

• Two Models were solved
• The total change in cost on a project
• The overall total cost of a project
• Based on the historical totals for RFIs and new
  drawings by division, and Austin initial
  forecasted budget

15
Two Models

• Using both models together
• Forecast total for RFIs and new drawings based
  on initial input of percent complete of the job
  and current totals of RFIs and new drawings
• Use those forecast to forecast the total change
  in cost of the project and overall total cost of
  project

16
Model Output

• Cubic Regression Models
• High R-squared terms
• Civil models had highest R-squared terms also
  had largest confidence intervals
• Architectural models had S-squared terms of 100
  - had the smallest confidence intervals
• Multiple Regression Models
• High R-squared terms
• Total RFI and Civil RFI variables seemed too have
  largest influence in both models
• Low probability that there terms where zero
• Total cost model more accurate that total change
  in cost model
• Significantly higher F-ratio and corresponding
  P-value (probability)
• Overall both models were strong and did good job
  of representing the data

17
Sample Output
18
Sample Output
19
Sample Output Continued
20
Recommendations

• Austin should use these models to help forecast
  RFIs and new drawings issued and their cost
  impacts
• Spreadsheet to allow Austin use these forecast
  models
• Austin should collect monthly RFI and new
  drawings issued information on all of its jobs
• Give current information to run forecasts with
• Provide historical data to add to current models
  to make more accurate
• Conclusion With the help of these models and
  Austin Commercials realization and efforts to
  solve this change management problem, I believe
  Austin Commercial can set itself apart from its
  competitors and better serve its customers in the
  long run.

21
Spreadsheet
22
Assumptions and Limitations

• Only uses data from last 3 years assumes last 3
  years is indicative of future
• Projects that were used for data ranged from cost
  of about 500,000 to 50,000,000 model may not
  be accurate for extremely large projects
• Assumes RFIs and new drawings issued are the
  best indicators for change in a project
• Amount of projects used - Would have like to
  included more projects in the data (hopefully
  more data will be available in the future)

23
Thanks and Questions