Used Trucks Are Trading Like a Volatile Stock. Most Dealers Don't Have a Chart.
A 2022 Ford F-150 XLT was a $42,000 truck in February 2025. By May 2025 it was $38,000. By August 2025 it was back to $41,500. By Christmas it was $37,200 again.
That's an asset moving 6 to 10 percent every quarter. If a stock did that, every desk on Wall Street would have a real-time chart, three indicators, and a risk model on it. The dealer trading that same vehicle gets a weekly Manheim email and a four-day-old MMR figure.
Today we're launching CarCast on Product Hunt. The pitch is one sentence: MMR looks back. CarCast looks forward.
The actual problem
Every used-car pricing tool a dealer uses today answers the same question: what did this vehicle sell for last week? MMR is a backward-looking national average. KBB is a retail estimate based on past transactions. Black Book is a wholesale guide that publishes weekly. vAuto layers your inventory on top of those.
In a flat market, that's enough. The market hasn't been flat for two years.
Three things happened in 2024–2026 that broke the backward-looking model:
-
EV residuals fell off a cliff. A 2021 Tesla Model 3 Long Range that was $42k in early 2024 was $26k by late 2025. Anyone who bought one at auction in Q1 2024 and held through Q3 ate a fifteen-thousand-dollar haircut.
-
Trucks went vertical, then mean-reverted. Wholesale truck prices climbed 12% in a single quarter in late 2024, then gave back almost all of it in the next two quarters. The dealers who bought the rip and held into the correction had inventory they couldn't sell at auction prices.
-
The collector market split from the daily-driver market. A 1988 Porsche 928 is up 19% on our forecast this week. A 1978 Ford F-150 — same vintage class, very different buyer pool — is forecast down 18%. "Classic cars" stopped being one category.
Each of those moves was visible in the listing data weeks before MMR caught up. The dealers who saw it first made money. The ones who didn't kept buying based on last week's average.
What we built
CarCast forecasts where used-car prices are heading 30 and 60 days out across 300+ vehicle segments. Each forecast comes with three things:
- A probability distribution, not a point estimate. P10, P50, P90 confidence bands so you can see the range of likely outcomes, not a single guess.
- A trend classification — Rising, Stable, or Softening — for at-a-glance scanning.
- Macro context — the Manheim Used Vehicle Value Index (wholesale) and the BLS used-car CPI (retail) on every segment, so you can sanity-check the forecast against the broader cycle.
Pricing for a single segment looks like this:
2022 Ford F-150 XLT
Current median: $38,858 (avg listing miles: 59k)
8-week forecast: $39,053
Expected change: +0.5% (Stable trend)
P10 forecast band: $38,400
P90 forecast band: $40,400
Confidence: 79%
Dealers don't get one number to second-guess. They get a chart with error bars, two macro indicators, and a directional read. Then they make their own call.
Why probabilistic forecasts beat point estimates
The single biggest mistake in forecasting any market is treating the median as the truth.
Take the 2022 F-150 above. The P50 forecast is $39,053. The P10 is $38,400 and the P90 is $40,400. That two-thousand-dollar spread is the model telling you: this segment is genuinely uncertain. Don't bet the lot on a $39k number.
Compare to a Honda Accord 2022, where our band is much tighter — maybe $300 wide. That's the model saying: this is a stable, predictable segment. You can lean on the median.
The width of the confidence band is itself a signal. Wide bands = handle with care. Narrow bands = confidence is high, act on it. That's information you simply can't get from MMR.
How the forecasts get made
Quick technical version, because some of you will care.
The forecasting engine is a 200-million-parameter foundation model for time-series forecasting, used unmodified. We feed it 52 to 104 weeks of weekly median price history per segment, plus the Manheim Index and the BLS CPI as exogenous covariates.
Pre-processing: log-transform the price series to stabilize variance, normalize against MUVVI to detrend macro effects.
Post-processing: extract the P10/P50/P90 quantiles, reverse the log transform, classify the median move as Rising/Stable/Softening at ±1.5%, score confidence based on band width and historical volatility.
The full inference batch — 374 segments, two horizons each — runs on a GPU cluster in under fifteen minutes every Sunday morning. Then it sits in Postgres until you query it.
If you want the deeper version, we wrote a longer technical post on the model.
What's actually moving this week
Some real numbers from the latest run, picked because they tell different stories:
The truck split. A 2013 Ford F-150 — well-depreciated, mid-mileage, classic daily-driver — is forecast +19.6% over the next four weeks. Strong used-truck demand at auction is dragging late-model prices up. But a 1978 F-150 is forecast -18.9%. Pandemic-era classic-truck bidders have moved on. Same nameplate, opposite directions.
Tesla's bifurcation. A 2023 Tesla Model Y RWD is currently classified Rising in the daily-driver/commuter pool. The 2020 Model 3 Standard Range Plus is Softening. The market is rewarding newer batteries and longer warranties; older EVs without those advantages keep depreciating.
The collector mid-stack. A 2008 Mercedes-Benz SL550 is forecast +19.2% this month. A 1988 Porsche 928 is +19.0%. A 1966 Cadillac Eldorado is +17.6%. The mid-six-figure collector market continues to find new buyers, particularly in segments with limited annual production.
These aren't predictions of certainty. They're probability distributions with informative confidence bands, backed by 52+ weeks of evidence per segment.
What this changes for you
If you're a used-car dealer, CarCast is the difference between buying at the auction lane based on Tuesday's MMR and buying based on a forward-looking forecast that includes this week's listing data, last month's wholesale trend, and macro context. The 60-day horizon is long enough to be useful for inventory decisions and short enough to actually be accurate.
If you're a wholesaler or fleet remarketer, the segment-level granularity means you can see the moment a class starts to soften before it hits the auction lane and gives you time to reposition.
If you're a retail buyer, you can stop guessing whether to buy now or wait. Pull up the segment, check the trend pill and the 8-week chart. If it's Rising and the band is tight, the conversation is short.
What we want from you today
CarCast is live on Product Hunt today: carcast.ai on Product Hunt.
We have a free plan with no card required. Pick the five vehicles you watch, see what the model thinks, tell us where it's wrong. We pay attention.
If you're a dealer, look at the Pro plan — $149/mo unlocks the 60-day horizon, alerts, and the weekly market digest. If you run inventory at scale, the Agency plan at $499/mo adds dealer-grade rate limits and priority support.
The model isn't right about every segment every week. Forecasting is hard, used cars are weird, and any system that claims certainty is lying. What we can tell you is the range of likely outcomes, the confidence behind each, and the direction the data is leaning. That's the chart most of the market doesn't have. Now you do.
— Rick
→ Try CarCast free → See current market trends → Find us on Product Hunt
Forecasts and trend classifications are informational analytics only and do not constitute financial advice or a recommendation to purchase, hold, or sell any vehicle. Individual market conditions vary.