Parametric Cost Estimating

Parametric cost estimation uses the statistical relationship existing between performance and physical factors associated with a project, and the actual historical costs incurred.  Performance and physical factors may include the fixed and variable costs associated with actual project performance, e.g. an engine fuel consumption and power output, contractor costs and value added or operational manning requirement needed. By combining predictive models to estimate what costs should be with real-world experience of what costs have been, a far more accurate estimate can be produced.


Parametric cost estimating uses a statistical regression analysis, and is most commonly utilized in the early stages of a project or product development; usually referred to as the System Development & Demonstration phase or SDD. It is widely used by government, military and industrial clients and their primary contractors, because whilst accurate costs are unknown due to a lack of detailed information (which only becomes available at the acquisition stage), the estimating solution is able to take into account top-level factors of design and performance.  In addition, the solution also provides risk analysis measures, for instance, probability of project success.

Effectively, parametric cost estimation is using two models: a forecasted projection and a historical model.  These are then combined to provide the project estimate.  It is therefore essential that the most current information is used in both models; i.e. the most recent historical cost information and the most up-to-date input and performance data. In addition, it is essential to compare like-with-like, so prior period estimations and data are standardized to take into account the time value of money, the impact of inflation in creating current pricing and the performance characteristics of the underlying technology.

An example would be a project requiring an estimate of the cost of a logistical fleet management solution, using historical information derived from a fleet management solution prior to the introduction of satellite tracking technology.  The cost estimate yielded will be a great deal higher than the real-world pricing, because the modern solution available fails to have the current cost-benefits of satellite tracking technology imputed into the estimation model.  A modern day example would be the estimation of a computer system using older cost and performance estimates to compare with.  Again, the cost estimates would be higher and the performance output estimates reduced, because of the advances in the underlying technology (computer speed, memory cost and increased functionality) bringing benefits and cost savings which would not be taken into account.

A major strength of parametric cost estimation is the ability to analyze how variable parameters affect cost and risk.  Cost Estimating Relationships (CERs) can be developed between the historical and forecast models, which can be manipulated to determine cost and performance changes if the underlying parameter is altered. Regression analysis is used to plot the CER’s impact, most frequently by using a “line of best fit” or in technical terms, the coefficient of determination.  This is why it is important that data collected to produce the CER is “normalized” in terms of time cost of money, but also in respect of performance and output characteristics of the underlying technology.  By varying the controlling parameter along the line of best fit, more accurate cost and performance estimates can be produced, provided the underlying CER database is operating within the real-world parameter values.  For instance, it would be inappropriate to generate a range of CERs for a family saloon car using NASCAR racing data, or to use CERs from gas-powered vehicles to generate cost estimates for hybrid fuel models.

In summary, parametric cost estimating provides significant advantages in cost estimating because it is using cost and performance data from different historical sources and from best-estimates of current data.  This generates a less risky estimate than one generated by analogy for instance.  Additionally, by using statistical techniques, the margin of error is better quantified which renders sensitivity analysis more meaningful.