Auto warranty and driving patterns
Introduction
Understanding the occurrence of warranty claims is of significant interest to manufacturers for two reasons: first, they are a liability incurred at the time of sale and represent the cost of doing business, and second, they allow engineers to assess the reliability of the sold products. Nowadays, the product warranty is considered as an attribute of the product and it is used by manufacturers as a competitive tool in the market place. For example, increasing warranty coverage may attract more buyers, but at the same time, could increase the related servicing costs. See Blischke and Murthy [1] for more discussion on these general points.
Typically, automotive warranties offer a free repair subject to age and mileage limitations. Usually the age of a vehicle is measured from the time of sale to the current time. In this study the limits for “bumper-to-bumper” coverage are 36 months or 36 000 miles, whichever comes first. Some manufacturers offer longer warranties across the board, others offer longer warranties only on their luxury models or for selected components such as powertrain, as for example, the Korean car manufacturer Kia.
Vehicles’ ages are known at all times because sales records are available. On the other hand, the mileage is available only for vehicles that generate warranty claims—this mileage is recorded at the dealership and included in the warranty database. Thus, from a modeling standpoint, we have two usage measures (age and mileage), but one of them (mileage) is incompletely observed. As is commonly done, here the warranty claims are modeled as recurrent events from a repairable system. Also, we use a nonparametric approach because our sample sizes are large. We note however that warranty forecasting, which requires extrapolation beyond the oldest age/mileage in the field, requires either a parametric model or the incorporation of past-model data on older vehicles. We deal explicitly with the double censoring in our dataset: the incomplete mileage information and the fact that claims made beyond the warranty database cut off date are not part of the database.
This modeling approach and data structure has been discussed extensively in the past. The model and estimation procedure are based on the ”robust estimator” discussed by Hu and Lawless [5], [4]. The consideration of the number of units at risk due to mileage limitations and based on incomplete mileage information extends Nelson [12] standard estimator. Lawless et al. [9] also dealt with the incomplete data problem and specify a simple linear mileage accumulation model, which is generalized in this study. The general survey paper by Lawless [8] includes a discussion of the bias caused by reporting delay, which was analyzed earlier in Lawless and Nadeau [10] and Kalbfleisch et al. [6].
Recently Rai and Singh [13], Majeske [11], Kleyner and Sanborn [7] and Wu and Akbarov [15] have proposed possible approaches to warranty prediction. Also, see [14] for more on reliability analysis with warranty data.
Chukova and Robinson [3] adopted the robust estimator and linear mileage accumulation model to estimate the number of units at risk at any given time from the incomplete mileage data. This information was used to provide explicit expressions for the estimated mean cumulative number of claims per vehicle with or without adjustments for reporting delay. Christozov and Robinson [2] relax the linearity assumption for the mileage accumulation, proposing instead a piece-wise linear model with nodes occurring at the observed claim times and mileage. In Chukova and Robinson [3] only the last warranty claim was used to estimate a vehicle's mileage accumulation rate whereas for the piece-wise model, all claims in the database were used to characterize the mileage accumulation. In Christozov and Robinson [2], using strata approach, the variability of mileage accumulation rates is defined and comparison between the mean cumulative functions based on this approach with the more basic one given in Chukova and Robinson [3] is provided. Also, an interesting conjecture, that a higher variability in the driving pattern may lead to a higher expected warranty cost (or higher expected number of claims) per vehicle, is formulated. This conjecture is the main focus of this study.
This paper sets out to test whether there is a relationship between the variability of the driving pattern of the vehicle and the frequency and size of the warranty claims made against it.
The content of this paper is as follows. Section 2 outlines the framework and the data that are used in this study. Section 3 introduces the idea of vehicle's “observable life”. In Section 4 the measure for the variability of the driving pattern is proposed. Results for the subset of vehicles with two or more claims are given in Section 5. Vehicles without claims are reintroduced in Section 6. Section 7 specifies the measures of warranty performance and Section 8 summarizes the results for these measures. The direction of causality is discussed in 9 Causality, 10 Conclusions concludes our study.
Section snippets
Intentions and definitions
Similarly to Christozov and Robinson [2], we assume that the vehicles' trajectories are piece-wise linear with nodes occurring at the observed claims' time and mileage. Consider a graph of mileage accrued by a vehicle against the age of the vehicle. The slope of this graph gives the rate of mileage accumulation per unit time. We refer to this graph and the slope across time as the driving pattern of the vehicle (see Fig. 1). We wish to find some measure for the variability (or volatility) of
Establishing a vehicle's observable life
For now, we consider only vehicles with claims. We calculate the mean rate of mileage accumulation for each vehicle, as specified in (1). This is then used as the best estimator for the usage of the vehicle, after its most recent claim, to help establish the observable life of a vehicle. Then, in the next section, we introduce and calculate a measure of variability of the driving pattern before we reintroduce vehicles without claims.
For our analysis, a vehicle is of interest as long as it can
Measure of variability
Next we introduce a possible measure for the variability of the driving patterns (VDPs) and discuss the effect of the vehicles' observable life forecasting on this measure.
We are interested in the variability of the vehicles' driving patterns as recorded in the database. Assume that a vehicle has warranty clams recorded in the database. One reasonable approach of detecting this variability is to measure the deviation of the driving patterns from the weighted average of the mileage
Results for vehicles with claims
We consider all vehicles with claims in the year 2000 dataset. The warranty records were sorted by variability of the driving pattern and graphed against the total cost per vehicle.
We remove from this graph all vehicles with zero variability, about 9096 vehicles. This is done for two reasons: discussing the variability of vehicles with a single record does not make sense and that it is ambiguous as to how these vehicles should be sorted in relation to each other, as a different arrangement of
Vehicles without claims
Manufacturers are interested in the number and cost of claims with respect to all vehicles sold. Recall that the analysis in the previous section is possible for only a subset of vehicles: those with two or more claims. Hence, now we reintroduce vehicles without warranty claims. Our goal is to calculate the mean cumulative cost to the manufacturer per vehicle, so we need to average over all vehicles including those with no claims.
Warranty measures
We now consider measures of a vehicle's warranty performance that could be useful to the manufacturer. These measures are calculated using the previously constructed grouping.
Results for all vehicles
Fig. 6, Fig. 7 show the adjusted and unadjusted measures for the expected cumulative cost and per vehicle and the expected cumulative number of claims and per vehicle. Each point gives the average value of the corresponding measure for the group, on the y-axis, against the median VDP of vehicles in that group, on the x-axis.
These average values are taken by calculating the measure for a certain group 50 times based on the vehicles with claims and one of the
Causality
All the preceding analysis is based on the assumption that the observed variability of the driving pattern is the cause of the higher number of claims and the higher cost per vehicle, but this may not be so.
Conclusions
This study suggests a model for extrapolation of the rate of mileage accumulation between the most recent warranty claims and the end of the vehicle's observable life. Also, we propose an appropriately constructed measure for the variability of the vehicle driving pattern.
Based on the proposed measure of variability, our results support the suggestion that the variability of the driving pattern affects the cost to the vehicle manufacturer of warranty claims made against the vehicle. The greater
References (15)
A non-homogeneous poisson process predictive model for automobile warranty claims
Reliability Engineering and System Safety
(2007)- et al.
Support vector regression for warranty claim forecasting
European Journal of Operational Research
(2011) - et al.
Product warranty handbook
(1996) - Christozov DJ, Robinson SC. Automotive warranty data: estimation of the mean cumulative function using stratification...
- Chukova S, Robinson J. Estimating mean cumulative functions from truncated automotive warranty data. In: Wilson A,...
- et al.
Estimation from truncated lifetime data with supplementary information on covariates and censoring times
Biometrika
(1996) - et al.
Estimation of rate and mean function from truncated recurrent event data
Journal of American Statistical Association (JASA)
(1996)
Cited by (13)
Aggregate discounted warranty cost forecasting considering the failed-but-not-reported events
2017, Reliability Engineering and System SafetyCitation Excerpt :To address this issue, Polatoglu and Sahin [7] studied the probability distributions of warranty cost that incorporate customer repurchase behavior for a failed unit, and Bai and Pham [8] derived the mean and variance of the discounted warranty cost for repairable systems with minimal repairs. Most of the research on warranty cost modeling only emphasizes the product and manufacturer related factors such as design reliability [9], maintenance strategies [10,11], and warranty policies [12,13]. One important factor that is often overlooked is the failure reporting behavior of a customer.
Reliable product-service supply chains for repairable products
2016, Transportation Research Part E: Logistics and Transportation ReviewCitation Excerpt :This paper fills this gap by incorporating forward and after-sales SCs. (III) Warranty cost: In these papers, authors focus on minimizing the warranty cost by scheduling appropriate maintenance (replace and repair) activities (Chen and Chien, 2007; Chukova et al., 2007; Williams, 2007; Wu et al., 2007; Wang et al., 2008; Rao, 2011; Vahdani et al., 2011, 2013; Su and Shen, 2012; Wang, 2012; Tsoukalas and Agrafiotis, 2013; Shahanaghi et al., 2013; Park et al., 2013; Anastasiadis et al., 2013; Liu et al., 2013). Rao (2011), for example, proposes a decision support model for repair and replacement decisions for a repairable product sold under a failure free warranty so that total warranty cost is minimized.
A two-stage preventive maintenance optimization model incorporating two-dimensional extended warranty
2016, Reliability Engineering and System SafetyCitation Excerpt :The assumption (5) is essential for developing customized PM programs. Previous literature usually bears the following consensus—the age is known for all sold items at any time (since the manufacturer generally knows the sale date), but the usage is only observed for those failed and claimed items [34,35]. However, in this study, due to the implementation of scheduled PM actions within the BW period, the maintenance crew can record the customers’ usage details (such as item age, usage, failure modes) at each PM instant, and these data can then be used to identify the customers’ average usage rates.
Construction of asymmetric copulas and its application in two-dimensional reliability modelling
2014, European Journal of Operational ResearchCitation Excerpt :There is considerable research on warranty data analysis. For more detailed discussion, the reader is referred to recently published review papers (Wu, 2012, 2013) and papers on reliability data analysis (see (Anastasiadis, Anderson, & Chukova, 2013; Wu & Akbarov, 2011, 2012), for example). However, the above methods have the following drawbacks.
Profit Optimization for Mileage-Based Pricing of Electric Vehicle Lease
2022, IEEE Transactions on Engineering ManagementField reliability estimation of agricultural tractors based on warranty data
2021, Transactions of the ASABE