A Novel Non-Recurring Production CER Methodology
Methods and Models I Track
Production costs are generally categorized as either non-recurring or recurring. Typically non-recurring costs include tooling and pre-production activities, among others. Cost estimating relationships (CERs) are generally developed first for recurring hardware costs, and then non-recurring CERs are developed as a function of recurring hardware costs.
Non-recurring hardware costs are notoriously difficult to estimate for two reasons. First, a particular contractor may not break-out non-recurring costs while others may break-out non-recurring costs using definitions inconsistent with other contractors. Second, non-recurring costs may only be relevant for certain productions lots such as very early lots, lots which represent a significant increase in production quantity, and lots which represent a change in design. Therefore, actual production cost data show a mix of production lots with zero and non-zero non-recurring productions costs.
A common way to develop a non-recurring CER is to calculate a factor as the ratio of non-recurring hardware cost to recurring hardware cost. Another way is to calculate the non-recurring hardware cost as a percent of the recurring hardware cost for each production lot and then regress these percentages as a function of the lot cumulative quantity. The non-recurring costs tend to be higher for early lots and are often zero for later lots, so a CER based on quantity will result in a steeply declining estimate as the quantity increases which better fits the data. If a large number of lots have zero non-recurring costs, both estimating methods will have large statistical errors, but the CER method will have smaller errors.
This paper provides a novel, alternate methodology to reduce statistical errors. This methodology combines (a) a method to predict whether the non-recurring costs will be zero or non-zero and (b) the CER regression method mentioned above, but with only the non-zero values of non-recurring hardware cost as a percent of the recurring hardware cost for each production lot.
The method to predict whether the non-recurring costs will be zero or non-zero is known as logistic regression. It is a technique used to find relationships between an independent variable that can take on multinomial, categorical values (i.e. binary or other multinomial value) and a series of dependent variables. Logistic regression has been used in studies of cost growth (Lucas & White, 2009) (White, Sipple, & Greiner, 2004) and failure analysis modeling to model bimodal (i.e. zero and non-zero) behavior of data.
In this paper we will (a) provide example data for weapon systems from which we will derive non-recurring CERs, (b) walk through the methods typically used to estimate non-recurring production costs and their weaknesses, (c) discuss the logistic regression technique, (d) show the logistic-regression-enhanced CER, and then (e) show the error analyses for the different methods of estimating non-recurring costs.
Ms. Hackbarth is a Cost Analyst at MCR, LLC. She has been involved with cost estimating for the Joint Urban Test Capability as well as the Missiles and Munitions Automated Cost Database for three years. Prior to working with MCR she was a Graduate Teaching Assistant for University of Central Florida’s (UCF) Department of Economics and a Multifunctional Financial Analyst at Lockheed Martin. Her specific experience covers cost estimating, defining program/project Work Breakdown Structures, and working manpower requirements. She received her BSBA in Economics and an MBA from UCF in 2008 and 2009 respectively.
Mr. Covert is the founder of Covarus, LLC, a boutique firm specializing in business, scientific and mathematical consulting. He has authored over sixty professional papers on the subject of cost and schedule analysis over the past twenty years, and is an internationally recognized expert in in these fields. In his twenty-six years in DoD weapon system and aerospace engineering he has also been employed at MCR, LLC, The Aerospace Corporation, Tecolote Research, Inc., LTV Corporation and several divisions of Northrop Grumman Corporation in the United States and the United Kingdom. He has specific experience in systems analysis, systems integration, parametric analysis, risk analysis, digital filtering, algorithm development, and simulation.