Published 2026-10-04 · LuckyMDM Blog
The short answer: a remarketing quote for one model in one cosmetic grade is not a number, it is a distribution. Treating it as a single figure biases every residual value forecast in the same direction - upward. Three parameters describe the distribution well enough to budget from: the median, the interquartile range and the sample size. Forecast on the median and never on the observed high. The reason the bias runs one way is simple: the high price you remember comes from a retail relisting channel, while the price you actually realise comes from a bulk channel, and those two channels are not quoting the same product.
A 128 GB handset graded as Grade A draws 400 USD from one buyer, 350 USD from a second and 310 USD from a third. None of the three is wrong, because none of the three is answering the same question.
The underlying mechanism is that a remarketing quote is not a measure of what the unit is worth. It is a conditional statement of what that buyer will pay, in that channel, on that specification, on that day. The buyer's downstream outlet sets the ceiling. A buyer who relists individually into a retail market at 420 USD can pay 400 USD and still hold margin. A buyer moving pallets into an export lot with 18 USD of margin per unit has a hard ceiling near 350 USD. The downstream structure of the channel determines the ceiling; the cosmetic grade only determines where inside that ceiling the quote lands.
This is also why collecting a single quote means treating three different products as one.
Recording one number is an implicit claim that the variance is zero. Three parameters are needed before the figure is auditable:
| Parameter | How to compute it | What it tells you | Common mistake |
|---|---|---|---|
| Median | Sort all samples ascending, take the middle one (average the middle two for an even count) | What you normally realise - budget on this | Using the mean, which one or two retail highs drag upward |
| Interquartile range | 75th percentile minus 25th percentile | How unstable the price is - stability matters more than level | Using the full range, which isolated outliers inflate |
| Sample size | Number of quotes and the collection window | Whether the first two mean anything | Calling three quotes a median |
With any one of the three missing, the distribution is still an impression rather than data.
Setup: one 128 GB model, Grade A, battery health 88 percent, one quote per day from three channels over 30 days, 60 samples in total.
Lowest 310 USD
25th percentile 336 USD
Median 350 USD
75th percentile 364 USD
Highest 400 USD
Interquartile range = 364 - 336 = 28 USD
Highest - median = 400 - 350 = 50 USD
50 / 350 = 14.3 percentForecasting on the 400 USD observation instead of the 350 USD median overstates the unit by 14.3 percent. On a fleet that remarkets 200 units a quarter, that is 10,000 USD of book value that does not exist. It rarely shows up as an outright loss - it shows up as a quarter that underperformed the plan, and nobody traces it back to the residual assumption.
Two quotes are not comparable until four fields are aligned. Miss one and any comparison is invalid:
Of those four fields, cosmetic grade and accessory completeness need a human. The other half is readable on the unit itself before it leaves the depot:
LuckyMDM (Sichuan Starlight Network LLC) exposes battery health, the seven-part hardware integrity states, the activation lock state and the enrolment binding state as fields read automatically before a unit goes to remarketing - the four items on a quote specification that are measurable on the device. Cosmetic grade and accessory completeness still require a person.
| Channel type | Level relative to median | Stability | Volume it suits |
|---|---|---|---|
| Retail relist | Above median, can reach the observed high | Low, moves with retail pricing day to day | Single units to a dozen |
| Bulk wholesale | Close to median | Medium, negotiated per lot | Dozens to hundreds |
| Consignment platform | Below median | High, but slow to settle | Odd units |
Forecasting from the observed high produces an error that only travels one way, because settlement almost always happens in the bulk channel rather than the retail one. Forecasting from the median produces an error that runs both ways and cancels over time. A forecast is there to reconcile against over many quarters, not to look best in one.
A great deal of effort goes into choosing the week to sell. In the sample above, the spread across channels - 310 USD to 400 USD, a 90 USD band - is wider than the drift within any single channel across the 30 days. Timing is not irrelevant: once a model's residual rate falls below roughly 42 percent of its launch price it tends to enter a faster decline, and that position is worth watching. But the channel structure sets the range, and timing only moves you inside that range.
Why it is wrong: the mean is pulled up by one or two retail highs, and settlement happens in the bulk channel. The median is insensitive to isolated high values, which is exactly the property this problem needs.
Why it is wrong: the low quote usually bundles freight, a reinspection clause and immediate settlement. It is a different bundle, not a worse price. Align the four specification fields before comparing.
Why it is wrong: cosmetics settle the appearance half only. Within one grade, a battery one band lower, a non-genuine display or an active Activation Lock each triggers its own deduction - and half of those are readable on the device before the buyer's reinspection decides them for you.
Boundary one: this page covers remarketing quote specifications and how to build a residual number. Acquisition cost allocation, instalment structure and deposit handling are a separate set of books and folding them in breaks the arithmetic above. LuckyMDM sits on the device side of that line - it reports device state, not commercial terms.
Boundary two: below 30 samples or a window shorter than 30 days, do not use the median either. With a small sample the median is simply whichever quote happened to land in the middle, and it moves when the collection window moves. Either extend the collection or label the figure an estimate rather than an observed value.
Yes, but the purpose changes. With one channel you are building a time series to measure drift. With several you are building a cross-section to measure channel spread. Either way, collect for at least 30 days.
Not by adjusting the number down. Record the volatility instead: models whose interquartile range exceeds 10 percent of the median go on their own line, so whoever reads the report can see where the uncertainty sits.
It is a channel cost only if you actually use that channel. The difference between a retail listing and your realised bulk price is not collectible in the bulk channel and should not be carried as an asset.
That depends on whether the buyer runs its own reinspection. Its value is not that it replaces reinspection, but that you know the result before the unit ships, instead of arguing about grading afterwards.
Yes. 128 GB and 256 GB need their own sample groups, because their distributions have different centres and a blended median is accurate for neither.