6 September 2023
The Market for EV Lemons: Battery Health and the Residual Value Conundrum
Accurate residual value prediction and battery health transparency are critical to the successful adoption of Electric Vehicles (EV). Confidence in residual values enables financiers to accurately monitor, predict, and control their balance sheet exposure, allowing them to pass through this reduction in risk in the form of more attractive credit prices and more widespread credit availability to prospective EV buyers. Transparency in battery health will empower the successful instantiation of a thriving second-hand EV market, providing consumers with the reassurance they currently lack regarding expected future capability of their used EV.
There is much speculation over the contribution battery health and expected future performance will have on EV residual values. Given the significant costs associated with battery replacements and the notable effects of degradation on asset performance we believe that battery health will eventually be a meaningful determinant factor in EV residual values.
Market players are currently struggling to identify standardised, understandable, repeatable and cost-effective methods that offer the transparency that customers are rightly demanding on battery health and expected future performance. Current solutions have significant flaws. They either rely on self-reported, inherently biased and sometimes wildly inaccurate health estimates from OEMs or, untenable intensive testing that would significantly shrink throughput rates for remarketers. Both methodologies also only assess the health of the battery today, limiting any opportunity to provide material reassurance on future battery degradation to consumers in the form of extended warranties or performance guarantees.
In this article, we analyse the current state of the wild-west-esque market for battery health checks. Using anonymised case study data of over 30,000 EVs we demonstrate the deep flaws present in the available market offerings today and show that cloud-based monitoring is the only solution that provides both accurate and transparent reporting on battery health without requiring time-consuming testing.
EV residual values present an enormous risk to financiers. The nascency of the EV market means that third party valuation services (e.g. CAP HPI) and internal teams can no longer rely on large proprietary market data sets they spent the last 15 years+ accumulating to predict residual values. The pandemic supply chain squeeze and corresponding used car price bubble, rapid release cycles of new EV models (and brands), continuous and unpredictable price changes and spiking inflation globally, are all contributing factors to the current market volatility, with many lenders experiencing severe markdowns in EV book values. Given enough time and data it seems fair to assume that dominant brands, models, and pricing strategies will emerge allowing EV residual value prediction to converge back to good-old-times of ICE-like depreciation curve modelling. However, a mere lack of market data is only part of the story when it comes to the new paradigm of EV residual value prediction.
Batteries are unique in the way in which they age. Their degradation is a complex interplay of multiple factors. Importantly, battery degradation is path dependent. This means that given two identical EVs (model, age, odometer etc), their future degradation is not just dependent on their health today (referred to as State-of-Health (SOH)) but also their entire usage history. We now need to consider questions like: Did this driver only fast charge their vehicle? How low did the previous owner let the State-of-Charge (SoC) go before they charged? What temperature and SoC has the vehicle been stored at between journeys? In other words, financiers, lenders and remarketers will no longer be able to rely on a simple odometer and age lookup table in a wear and tear guide to assess vehicle condition.
But what does this all mean for residual values? The battery in an electric vehicle is the most expensive component by quite a margin. Some estimates put the cost of the battery as high as 45% of the total vehicle bill-of-materials, making replacements prohibitively expensive. The battery also defines the capability of the asset: degradation in the battery can have significant impacts on operationally significant parameters like range, performance, and charging speeds. No owner wants to take on the enormous risks of unpredictable performance degradation, and a potential costly maintenance, without the ability to hedge through extended warranties, performance guarantees etc. This lack of trust in future capability is often cited as one of the largest concerns for prospective used EV buyers, particularly in commercial applications. Given the high cost and criticality of the battery, it seems fair to assume that price volatility will continue until lenders can provide transparency and guarantees to consumers on the future performance of their used EVs. This market dynamic was first noted in the used ICE vehicle market within George Akerlof’s 1970 seminal paper “A Market for ‘Lemons’” where information asymmetry between dealers and consumers on the vehicle condition led to adverse selection, ultimately harming consumers and the trust in the used market.
What isn’t clear, is how sellers can provide transparency in a standardised, understandable, repeatable and cost-effective way. An OEM’s estimation of the SOH is often deliberately hidden from the end user, some third-party “quick-check” dongles exist that read this value directly from the battery, but many OEMs have significant flaws in their calculation of SOH (or maybe deliberately overestimate given their direct incentivisation to report positive results). Many EV owners have reported values of 100%, even after several years of use while others have reported volatile fluctuations in the reported health. Elysia analysis has found that industry-typical EVs can misreport SOH by up to 9%, with this error varying significantly from model to model.
A reliance on SOH as a single metric to describe future capability is also flawed. SOH provides a snapshot of the loss in battery capacity at a single point in time. Given that we know battery degradation is path dependent and sometimes highly non-linear, a single metric describing battery health today doesn’t tell us much about the future trajectory. Instead, battery engineers discuss battery degradation in terms of “Remaining Useful Life” (RUL), which describes the expected time of useful operation from a battery system until it reaches its end-of-life threshold (often set as 70-80% SOH in automotive applications). If sellers want to provide reassurance to consumers on future battery performance a SOH metric must also be paired with predictions for RUL for a range of use cases.
Regulators around the world have started to introduce frameworks that mandate SOH reporting requirements from battery systems. However, these aren’t due to come into effect until the end of the decade, and even then, the accuracy demanded is large enough that consumers will still be left with significant uncertainty.
The EU council recently voted to adopt new requirements on data reporting from batteries that requires manufacturers of vehicles, stationary storage and e-mobility products to ‘store the information and data needed [on the Battery Management System] to determine the state of health and expected lifetime of batteries. The legislation comes into force in 2026, and will require compliance within 12 months (i.e. by 2027). The UK has also been working with the United Nations Economic Commission Europe (UNECE) to develop similar regulations mandating SOH reporting from OEMs and minimum degradation performance targets on battery systems.
However, policy discussions on battery health are currently limited to warranty periods, with the EU, UK, United Nations Global Technical Regulation (GTR) and California Air Resources Board (CARB) covering only an 8-year period in lifetime and lifetime reporting. Sending a clear signal to manufacturers that it is acceptable for a battery to fail immediately after the 8-year period, despite EV lifecycle emissions assessments typically assuming a 15 to 20-year life and several hundred thousand kilometres.
The legislation distinctly separates State of Health (SOH) from Remaining Useful Life (RUL). While SOH gives information about the battery performance today, RUL enables used car buyers to make informed decisions about long term purchases and gives fleet owner-operators the data required to justify whether it is more economical to recycle or repurpose their battery assets.
Accuracy requirements are also insufficient in both fitness for purpose and timelines. CARB explicitly require a 5% SOH accuracy by 2026 which is broadly consistent with proposed UN GTR requirements. This level of accuracy is not adequate. Depending on the degradation pathway, a 5% SOH error can be equivalent to over a quarter of the asset’s life, and therefore is not a suitable target for deriving a meaningful remaining useful life estimate or distinguishing between self-limiting and knee-point trajectories (see Figure 1).
So, from a regulatory perspective, it’s too little, too late; OEMs are not obligated to supply SOH and RUL information until 2027, and even then, the required SOH accuracy is inadequate for accurate RUL prediction. This has led to a demand for independent, accurate battery health evaluation. But with many new players, how can we know who to trust? There is a clear incentive to oversell the convenience of a rapid test, at the (somewhat intangible) expense of accuracy. Some claim to measure the health of an EV battery within minutes, using dubious methods and devices. Others hide behind terms like ‘AI’ and ‘machine learning’ to avoid explaining their approaches, which ultimately undermines the founding goal of transparency.
In response to the market pull for a rapid, independent health check, one company claims to provide an assessment from a single vehicle acceleration event, in which voltage, current and temperature signals are used to estimate the relative change in the battery’s internal resistance compared to an expected condition at beginning of life. On the face of it, for the purpose of estimating the battery resistance (or efficiency), this method has some justification. However, resistance estimation can be highly volatile due to factors such as cell-to-cell variation and uncertainty in measurements (e.g. cell temperature). For these reasons, accurate resistance estimation generally requires many point estimates over a period as part of a statistical analysis. Accuracy concerns aside, the major flaw comes from the claim that battery capacity can be inferred from this resistance measurement, alone. This assumption is simply incorrect in most cases.
To illustrate this, Figure 2 shows Elysia analysis of data from a real automotive fleet of tens of thousands of vehicles, spanning approximately 160,000 km. While capacity drops steadily, internal resistance improves within the first 20,000 km as the battery is ‘broken in’, for reasons unrelated to normal capacity degradation mechanisms. Beyond 20,000 km, internal resistance increases gradually, but marginally. As a result, there is a clear one-to-many mapping from resistance onto capacity, and counterintuitively the battery resistance at end-of-life remains similar to the value at beginning-of-life, despite a significant capacity loss. In other words, a resistance measurement is not sufficient to infer capacity and, even in the monotonic region, the sensitivity of resistance change onto capacity is an order of magnitude too low for any useful insights to be derived. For the purpose of supporting vehicle remarketing at year 3-5 therefore, this approach is essentially inferior to valuing the vehicle on odometer and time in service, alone.
It gets worse; vehicle remarketers desire instant results, and are desperate to avoid lengthy controlled test drives, or plug-in charges. The market has responded, with one company doubling down on convenience, claiming to offer a health certificate within minutes without any test drives which involves switching on electrical loads within the vehicle such as the HVAC system, in order to draw a load. While this may appear credible, even a cursory assessment debunks its efficacy. The maximum plausible power from these auxiliary loads on a typical EV is approximately 10kW, which represents a slow discharge rate compared to ordinary driving. Not only is the voltage drop negligible from such a small power draw, further reducing the signal-to-noise ratio, but there is also a new fundamental limitation from the electrochemistry.
The conclusion, so far, is that none of the solutions discussed above address the brief for an independent and accurate health assessment. However, it is possible to independently verify battery health, and there are some services on the market that, on the face of it, could plausibly address this challenge. A true evaluation of battery health requires, at a minimum, an extended test in which the vehicle is typically instrumented with a physical OBDII device and subjected to a long drive that spans a very broad state of charge window, or a planned slow charge. In either case, the diagnosis time is on the order of hours or days and requires prior planning; it is therefore of little interest to the quick-turnaround demanded by used car sales platforms. It also fails to properly address the real issue for vehicle owners; the Remaining Useful Life. The current battery health is only part of the concern – the health trajectory is in fact more relevant to the residual value than the health itself. Battery degradation is complex, multi-dimensional and path dependent. For example, a vehicle that has been rapid charged regularly may have a significantly shorter remaining useful life expectancy than a vehicle that has not, even if both vehicles have the same battery capacity, today.
The reality is that the market cannot currently offer a solution that is both rapid enough to be compatible with the processes that remarketers are used to, and accurate enough to provide useful insight in setting residual values. This poses questions around motives – while the potential conflicts of interest for OEMs in reporting battery health are widely discussed, the same vested interests could just as well apply to remarketers, who in the end, only need enough to convince the (relatively uninformed) buyer of a battery’s good health. We would go so far as to state that the fledgling ‘battery health certificate’ market today is at risk of imposing net-damage to the automotive remarketing industry if not well regulated, as early movers jump on the bandwagon to favour the tangible benefits of ‘speed’ and ‘cost’ over ‘accuracy’ (which isn’t easily challenged). It’s hard not to draw parallels to Elizabeth Holmes’ Theranos, a company that infamously offered the ‘tiny blood test’ promising faster, cheaper and more convenient medical diagnoses, at your local pharmacy as opposed to hard-to-reach doctors’ offices. Moving forward, appropriate regulatory endorsement on methods and validation testing will be key. One thing is clear – ‘instant health check’ solutions have no place in the future of battery health certificates, and the question is whether the market wakes up to this before or after the damage is done.
What should the future look like? With advanced Battery Analytics, it is possible to identify and distinguish between the favourable and unfavourable degradation pathways and therefore, to generate lifetime projections that can be incorporated into a battery health certificate. However, this requires access to historical battery data, which calls for data connectivity to be planned in from the outset. The most obvious source for the data, to support scalability, is from the OEM, direct, as it can be sourced retroactively, and simply analysed at point of resale. While connected vehicle data can be purchased, OEMs generally remain protective over selling the signals required for an independent health assessment.
Moving forward, as the market for OEM-direct Data-as-a-Service grows, there is an opportunity for automakers to differentiate themselves by making available the signals that support independent battery health assessments. Tesla are already paving the way with their recent release of Fleet Telemetry, a ‘decentralized framework that allows customers to create a secure and direct bridge from their Tesla devices to any provider they authorize’. For the OEMs that stand firm, the business models for used car sales may simply evolve. For example, digital used car sales platforms could offer resale premiums to prospective vehicle sellers willing to install a telematics device for a period preceding the sale (e.g. a few hundred kilometres). Lease companies and fleet operators will plan the provision of appropriate telematics from the outset in anticipation for eventual resale, or simply choose the automotive brands with more open data policies; at which point, perhaps it’s just a matter of time before the OEMs converge on following in Tesla’s footsteps by providing the data, directly.
 Economic Commission for Europe. (GRPE) Proposal for a new UN GTR on Invehicle Battery Durability for Electrified Vehicles. United Nations Economic Commission for Europe https://unece.org/transport/documents/2021/12/working-documents/grpeproposal-new-un-gtr-vehicle-battery-durability (2022)
 California Air Resources Board. Final Regulation Order for Section 1962.8 Warranty Requirements for Zero-Emission and Batteries in Plug-in Hybrid Electric 2026 and Subsequent Model Year Passenger Cars and Light-Duty Trucks. https://ww2.arb.ca.gov/sites/default/files/barcu/regact/2022/accii/acciifro1962 .4.pdf (2022)
 Tankou, A., Lutsey, N. & Hall, D. Understanding and supporting the used zeroemission vehicle market. (2021)
 The first step along the road to transparency is to, at least, read the OEM’s internal State of Health estimate from the vehicle. While a small number of companies that offer a battery health certificate on this basis, any independent evaluation requires battery signals to be recorded while an appropriate electrical load is applied.
 Li-ion battery resistance dynamics are highly rate dependent at these low discharge rates because of fundamental limitations in reaction kinetics; drawing a consistent power every time or building a reliable normalisation surface to manage this dependency is nigh-on impossible.
 Despite this, one company attempts to offer a solution from OEM-sourced data but does so by misrepresenting a health expectation (given vehicle-level usage) for an independent health measurement. The service offers a ‘Range Score’ that compares the vehicle’s current maximum range to the original range for the same make, model and battery pack configuration, using only ‘odometer’, ‘state of charge’, ‘charging status’ and ‘range estimate’ signals.
Abstract Models are essential for engineering efficient systems both quickly and cost-effectively. They focus the mind on what’s important and inform where our precious testing resources should be allocated. Batteries are complex, with behaviours covering a wide range of length-scales and timescales; yet a strong understanding of them pays dividends at all stages of the […]
19 April 2023
Introduction Hardly a day goes by without artificial intelligence grabbing the headlines. Be it in process automation, medical diagnosis, autonomous transportation, language processing or even gaming, applications of advanced machine learning algorithms are ubiquitous. Outside of highly publicised advances such as ChatGPT, machine learning has equally spread across all branches of science and continues to […]
19 April 2023
Batteries are increasingly becoming a foundation on which society is built. Whilst the modern lithium-ion battery was largely developed to serve consumer electronics, its potential to fundamentally transform mobility and future green energy systems has led to something of a ‘second advent’ in this space. However, with this opportunity for battery technology comes challenges. Much […]
19 April 2023