On February 28, 2026, the United States launched airstrikes against Iran. Oil futures spiked. The dollar moved. Cable news talked about supply chains. And within days, a narrative started forming in aviation circles: this would change everything.
We decided to check.
At Windsock AI, we have a valuation model that can price every aircraft individually based on its specific characteristics. So when the war started and the market chatter kicked off, we had the data to ask a simple question: has the war changed the market at all yet?
The answer, after five progressively rigorous analyses across nearly 4,000 aircraft, is clear: the headline story of some apparent price shift is a composition effect. Through late March 2026, the war has not yet caused a meaningful repricing of the aircraft market.
A caveat upfront: we’re less than a month into what could be a seismic geopolitical event. There’s plenty of time for real pricing effects to develop. But as of right now, we’re not finding one—and folks claiming otherwise are likely being fooled by the statistical trap that catches everyone in thin markets like general aviation.
Here’s how we know.
If you follow aviation market commentary, you’ve probably heard some prognostication of some major price shift in the market, either already underway or imminent. The reasoning seems straightforward: you can, after all, just look at average listing prices before and after and event, see a difference, attribute it to the event. Simple.
The problem is that aviation is a thin market. Unlike equities, where millions of shares of identical stock trade every day, the aviation market moves a small handful of unique, non-fungible assets per month. No two aircraft are the same. A 1998 Bonanza with 3,200 hours is not the same as a 2004 Bonanza with 1,100 hours, even though both are “Bonanzas.”
This creates a well-known statistical trap called a composition effect. If the mix of aircraft listed in March happens to skew toward newer, higher-value types compared to February, the average price goes up—even if no individual aircraft changed in value. The market looks like it moved, but it didn’t. The listings just rotated.
Composition effects are the single most common source of false signals in thin-market analysis. They fool simple averages. They fool basic regressions. They even fool experienced market participants who are pattern-matching on anecdote. The only reliable way through them is a model powerful enough to price each aircraft on its own characteristics, so you can separate what the inventory is doing from what the market is doing.
We’ve written about this before in the context of why our valuation methodology is fundamentally different. The same problem applies here—and the Iran shock turns out to be a textbook case.
Let’s start with the simplest possible analysis. Take aircraft listed right before the war (February 2026). Then, take aircraft listed after (March 2026). Compare the averages.
That’s the headline number. Post-war listings averaged about $21,000 more than pre-war listings. If you stopped here, you could build a whole narrative about war-driven price increases. It sounds plausible. It matches the vibes.
But the p-value is 0.58. That means there’s roughly a coin-flip chance you’d see this difference even if the war had zero effect. The naive comparison can’t distinguish real price movement from inventory rotation. It’s exactly the composition-effect trap—and it’s where most market commentary stops. The market "moved" 3% in a single month! How could we be wrong?
A more careful approach would add controls. Aircraft age, engine hours, interior and exterior condition, segment type, whether it’s single or multi-engine—the standard variables that explain price differences between aircraft. If we "control" for the inventory, then whatever's left has to be the effect of the start of the intervention, right?
We ran this regression on log-price with a post-war indicator variable and basic aircraft controls. What we found may surprise you:
Now it looks significant, and an even stronger upward market effect. After controlling for basic characteristics, there appears to be a post-war premium of roughly 9.6%. This is the kind of result that looks rigorous enough to publish. It has a tiny p-value. It uses real controls. A reasonable analyst with standard tools would feel confident in this number.
But here’s the problem: an R² of 0.57 means these controls only explain about 57% of the variation in aircraft prices. The other 43% is unaccounted for—and in a thin market, that unaccounted variation is exactly where composition effects hide. Age and hours and condition scores don’t fully capture the difference between a well-equipped late-model turboprop and a basic trainer from the same era. The post-war indicator absorbs whatever the controls miss, and even a "sophisticated" market analyst would come away with a totally incorrect dramatic headline.
This is the dangerous result. It looks right. It passes conventional significance tests. And it’s wrong—because basic regressions aren’t powerful enough to neutralize composition effects in a market this heterogeneous.
This is where a real model changes the picture.
Windsock’s ML model prices each aircraft individually based on over 1,300 make/model combinations, accounting for age, accidents, hours, engine time, condition, avionics, location, macroeconomic conditions, market timing, and dozens of other factors that drive market value. When we use the model’s estimate as the baseline and test whether a post-war indicator adds anything on top of it, we’re asking a much sharper question: after explaining nearly all of the price variation, is there still a war effect?
The effect collapses. From 9.6% down to 1.8%, and the p-value crosses above the conventional significance threshold. The model explains 91% of the variation in observed prices—leaving almost no room for a war indicator to pick up residual composition noise. And with that better baseline, the apparent post-war premium largely vanishes.
This is the critical finding. The basic-controls regression found a “significant” 9.6% effect because it couldn’t fully separate real pricing from inventory mix. Once we anchored on a model that actually understands aircraft prices, the effect shrank by more than 80% and lost statistical significance.
The gap between Step 2 and Step 3—between an R² of 0.57 and an R² of 0.91—is where the composition effect lived. It’s the 34 percentage points of unexplained variation that the basic model couldn’t handle, and it’s exactly the kind of thing that generates false market narratives.
Before drawing any final conclusions, we need to ask a harder question: did our model break during the shock?
If the war disrupted the fundamental relationship between aircraft characteristics and prices—say, if buyers suddenly became irrational or if certain aircraft types became uniquely desirable for geopolitical reasons—then our model’s accuracy would start to diverge across the shock boundary. The model would be confidently wrong, and our Step 3 result would be unreliable.
We tested this directly. We looked at our model's accuracy error for the last two months, and tested whether that gap changed after the war started.
No statistically significant divergence. The model’s errors before the war look essentially the same as its errors after the war. Our model tracked prices just as well in March as it did in February. The shock didn’t break the pricing relationship.
This matters enormously, because it unlocks the final step.
Here’s the payoff. Because the model stayed calibrated through the shock, we can run a counterfactual that directly isolates the war’s effect on the exact same aircraft.
For aircraft on the market in February, we asked: what would our model have predicted if this same aircraft had been listed in March instead? And for every March aircraft: what if it had been on the market in February?
This holds inventory perfectly fixed. Same aircraft, same characteristics, same model. The only thing that changes is a small date change that just happens to jump a border of a big shock to the market. If the war caused a real repricing, the model would predict different values on either side of the shock boundary—and that difference would show up here.
Not positive. Slightly negative, and small enough that “approximately zero” is the honest characterization. The war didn’t cause aircraft to be priced differently. The model sees no repricing across the shock—just the same aircraft, valued the same way, regardless of which month you place them in.
Here’s what happens to the apparent war effect as we layer on better controls:
Step 2 is the trap. It has a tiny p-value and it looks rigorous. But the composition effect is hiding in the 43% of variation that basic controls can’t explain. The moment you bring in a model that actually understands the market (Step 3), the effect evaporates.
This is a textbook demonstration of why composition effects are so dangerous in thin markets. They survive basic controls. They produce statistically significant results. They reliably generate confident, wrong conclusions. The only way to reliably neutralize them is with a model that explains enough of the variation—well above 80%—that there’s nowhere left for compositional noise to hide.
We’re less than a month into this. A conflict of this scale could absolutely affect aircraft pricing over a longer horizon—we’re not claiming it won’t. What we’re saying is that through late March 2026, it hasn’t, and the early claims that it has, in any direction, are almost certainly composition effects rather than real repricing.
It’s too early to know whether the war will affect transaction times. Days-on-market data takes months to develop meaningful signal, and we’re only a few weeks in. If the conflict is creating buyer hesitation or a bid-ask standoff, that would show up as lengthening sale times before it shows up in prices—and we’ll be watching for it.
On the fuel side, oil prices have moved meaningfully since the airstrikes started. If elevated fuel costs persist for several months, there’s a plausible mechanism for downstream pricing effects: higher operating costs could compress demand for fuel-intensive types, particularly older jets and turboprops not on managed maintenance programs. That’s a real channel, but it operates on a longer timeline than four weeks.
What we can say with confidence right now: listing prices have not changed in response to the conflict. Any apparent movement is a composition effect—different aircraft came to market, not different prices. Anyone making buying or selling decisions based on a perceived post-war price shift is reacting to a statistical illusion, not a market signal.
A naive comparison says aircraft prices went up $21,000 after the war started. A basic regression says the effect is almost 10%. A model that actually understands the market says it’s roughly zero.
The difference isn’t a matter of opinion. It’s a matter of how much variation you can statistically explain. In a market as thin and heterogeneous as aviation, the gap between “looks significant” and “is significant” can be enormous, and is what separates opinions from facts. Composition effects fill that gap with false signal—and anyone drawing conclusions from simple before/after comparisons right now is almost certainly falling for them. The only way to cut through composition noise is with a sophisticated model capable of reasoning about what each individual aircraft is actually worth, not what the average listing happens to be this month.
This story isn’t over. If a real pricing effect develops over the coming months, we have the tools to find it. So far, it hasn’t.
