Progressive, and the Pool That Is Shrinking

We initiate on Progressive at Neutral, and it is the first Neutral we have published. This is the best underwriter in American personal auto, compounding at a 34.9% return on equity with a combined ratio of 87.3 against its own 96 target, and the equity has still fallen 11.3% over twelve months at 11.0 times trailing earnings. The de-rating is not a mistake. Average premium per policy is now falling, consensus has 2026 earnings declining, and the hard market that produced these margins is over. What the tape has not priced, in either direction, is that Progressive is the large insurer most exposed to autonomy, through a book almost nobody associates with driverless technology.

The near term is a cycle story and the market has it broadly right. Policies in force grew 7% in the second quarter while net premiums written grew 5%, and in June alone written premium grew 3% against the same 7% policy growth. Progressive is buying volume with price. That is a deliberate and well-practiced move at 87.3, with nine points of margin between the reported combined ratio and the 96 the company runs to, but it does mean the earnings power on display is cyclical rather than structural. The June combined ratio of 90.0 against 86.6 a year earlier is the first clear print of the turn.

The longer question is the one we came to test. Our Kodiak note argued that driverless freight rewrites commercial auto liability and that data-rich incumbents, Progressive named first among them, would be best placed to price the risk that replaces it. Working through Progressive's actual book, that view does not survive contact. Commercial lines are 13% of net premiums written, and inside that the company writes one to twenty truck for-hire operators, which is the last segment of freight autonomy will reach. The 87% of the company that is personal lines is where the autonomy exposure genuinely sits, and it is exposure rather than opportunity.

We are not negative on the business. Nothing in the data suggests the franchise is impaired, the telematics advantage is real and measurable, and at 11 times earnings the price is not demanding. We are Neutral because the asymmetry is absent at this quotation: the cyclical downside is live and identified, the structural downside is real but a decade out, and the upside requires the rate cycle to turn back sooner than the evidence supports.

Progressive wrote $83.2B of net premiums in 2025 across 40.1 million policies. The scale is not the interesting part. The mix is.

Net premiums written by segment, 2025

Segment shares from the 2025 Form 10-K. Property coverage sits inside both segments and is not broken out separately in this split.

Policy growth in the second quarter was concentrated where the company has been spending: direct personal auto grew 10% and agency personal auto 8%, while commercial lines grew 3% and property 1%. The company is putting its capital and its advertising behind the personal auto franchise, which is both its historical strength and, as Section 08 sets out, the book with the longest-dated structural question attached to it.

Policies in force, 30 June 2026

Commercial lines are 3.1% of policies and roughly 13% of premium, the gap reflecting a higher average premium per commercial policy.

The single most useful number in the current disclosure is not the combined ratio. It is the gap between policy growth and premium growth.

In the second quarter, policies in force rose 7% while net premiums written rose 5%. In June, policies still rose 7% while written premium rose 3% and earned premium rose 2%. Premium per policy is therefore declining, and the rate of decline is accelerating through the quarter. That is what the end of a hard market looks like from inside the income statement, and it arrives before it shows up in the combined ratio.

The combined ratio is now following. Second quarter 2026 came in at 87.3 against 86.2 a year earlier. June alone was 90.0 against 86.6, a deterioration of 3.4 points, and June net income fell 31% year on year even as premiums grew. Consensus has caught up: the 25 analysts covering the name forecast 2026 revenue down 0.4% and earnings per share down 1.5%, from $18.25 to $17.97, and the rating sits at Hold.

There is a benign reading and we take it seriously. Progressive runs the business to a 96 combined ratio and treats everything below that as margin available to buy growth. At 87.3 the company has roughly nine points of underwriting margin to spend, and it is visibly spending them: conversion is at its highest in more than twenty years, direct auto policies grew 10%, and the newest Snapshot pricing model has been extended to fourteen states representing 44% of trailing twelve month premium. Buying share at the top of the margin cycle is precisely the playbook that built the franchise.

The less benign reading is that competitors have recapitalised and the pricing environment is normalising regardless of what Progressive intends. Both readings produce the same reported numbers for several quarters. The distinguishing evidence is whether the policies added in 2026 retain and run at target loss ratios into 2027 and 2028, which is not observable yet. Until it is, we decline to underwrite either interpretation, and that is a large part of why the stance is Neutral.

There is no crowd here in either direction, which is unusual for a name we cover and worth stating plainly.

Short interest is 5.98M shares, 1.03% of a 579.1M share float. For comparison, the two names already under Contra coverage carry 19.47% and 6.87% of float short. Progressive is not a contested position, there is no squeeze setup, and no bear thesis is being expressed through the borrow. The consensus rating is Hold across 25 analysts with a $232 target, roughly 6% above the current quotation. The Street is neither defending nor attacking the stock, which is consistent with a view that the fundamentals are understood and the cycle is doing the work.

That absence matters for how the position should be sized. Our other two notes describe setups where positioning itself is part of the opportunity. This is not one of those. Any return here has to come from the underwriting and the multiple, with no help from a forced buyer or a trapped short.

Progressive launched Snapshot in 2008 and was first to usage-based pricing at scale. Seventeen years of driving data is a genuine asset, and unlike most claimed moats in insurance it produces an observable output: the company prices risk more finely than competitors and therefore selects better risks at the same premium.

The measurable evidence in the current disclosure is the deployment of the newest Snapshot model across fourteen states representing 44% of trailing twelve month net premiums written, which the company credits with improved segmentation and risk selection, alongside the highest conversion rates in more than twenty years. A first quarter combined ratio of 86.4 against a 96 target is the result that matters.

The advantage is in selection and pricing of human drivers. That is worth restating, because it is the specific thing our house view assumed would transfer to autonomous risk, and the transfer is not automatic. A model trained on how humans brake, corner and accelerate late at night has no predictive content for a software driver whose behavior is deterministic, versioned, and identical across every vehicle running that build. The actuarial problem for an autonomous fleet is not driver heterogeneity. It is software release risk, sensor failure modes, and manufacturer product liability, none of which appear in seventeen years of Snapshot data.

The moat is real, it is durable against other insurers, and it is currently pointed at a pool of risk that autonomy is designed to eliminate.

This is the section that changed our mind, and it turns on a distinction the sector conversation routinely skips.

Progressive is among the top three commercial auto insurers in the United States by direct written premium, which is the fact usually cited when the company is named as a driverless beneficiary. The composition of that book is where the argument fails. Progressive Commercial writes one to twenty truck for-hire operations. It is built for owner-operators and small fleets, it accepts brand new operating authorities that most carriers decline, and it competes on same-day filing and online quoting. It is a generalist with a trucking division, not a trucking specialist, and it does not write hazmat, household goods, heavy haul or autohaul.

Where driverless freight arrives, against where Progressive writes

Our Kodiak work established that long-tail carriers are 44.8% of carriers and 72.3% of crashes. That population is Progressive's customer. It is the part of freight with the worst loss experience, the least capital, and the longest runway before an autonomous tractor is economically available to it. Whatever else that book is, it is not a claim on the autonomous freight transition.

Our Kodiak initiation, published 5 September 2026, argued that the liability regime change in freight would advantage data-rich incumbents, and named Progressive first. Having done the work on Progressive specifically, we are amending that.

The mechanism in the house view holds. A driverless truck moves the insured event from the trucking company's commercial auto policy to the manufacturer's product liability, the legacy pool tied to human drivers shrinks, and a new pool forms around product liability, technology errors and omissions, cyber and fleet-level autonomous coverage. None of that changes.

What we got wrong was the entity. The new pool forms in large-fleet commercial and manufacturer product liability, which is written by the specialist and excess markets rather than by a small-fleet generalist. On the evidence in Section 06, Progressive has no franchise in either. Berkshire through National Indemnity remains a credible home for large unusual liability, and Markel through State National remains the fronting platform a new autonomous program would ride. Progressive does not belong in that group on the strength of its telematics alone.

The corrected position: within the trucking insurance complex, our preference order is unchanged for Markel and Berkshire, the paired-trade framing on Old Republic and Travelers stands, and Progressive moves from the data-moat beneficiary column to a separate question entirely, which is what autonomy does to personal auto. The Kodiak note's allocation map should be read with that amendment attached. We will restate it in that note's next revision.

If 87% of Progressive's premium is personal lines, and personal auto premium exists because humans crash cars, then the relevant question is what happens to that pool as cars increasingly do not crash. The honest answer is that the erosion is real, directional, and far slower than the technology headlines imply, because two forces work against each other.

The Insurance Institute for Highway Safety and its Highway Loss Data Institute have measured it directly on sensor-equipped vehicles already on the road. Collision claim severity runs about 10% higher on equipped vehicles, and property damage liability severity about 15% higher, because the sensors, cameras and calibration that prevent crashes are expensive to repair when a crash happens anyway. Frequency falls by more than that. The net effect is overall collision losses about 5% lower on equipped vehicles.

The offsetting forces, measured

A 5% loss reduction per equipped vehicle generation, phased in across a fleet that turns over across roughly a decade and a half, is not a cliff. It is a slow compression of the addressable pool, and it is partly self-limiting because more expensive vehicles justify higher premiums per policy even as accidents become rarer. This is why we treat the autonomy question as structural and long-dated, sitting outside the 12 to 24 month horizon this note carries.

The step change would come from liability transfer, not from frequency. If personal vehicles reach a level of autonomy where the manufacturer rather than the owner carries the liability, the premium does not merely shrink, it moves to a different balance sheet and a different product. That is the tail risk in the name, and its timing turns on regulation and litigation more than on engineering.

One observation we could not fully verify and are flagging rather than asserting: we did not find autonomy identified as a named risk factor in the summary of the 2025 Form 10-K that we reviewed, which listed competition and cybersecurity among principal risks. Confirming whether the full filing addresses autonomous vehicles, and in what terms, is the first item on the analyst agenda. If the largest personal auto insurer in the country is not naming it, that is itself informative about how far from the market's mind this sits.

Progressive's competitive problem in the near term has nothing to do with driverless technology. It is that the hard market repaired everyone else's balance sheet too.

Through 2022 and 2023, personal auto carriers took large rate increases to recover from severity inflation. Progressive came out of that period with the best margins in the industry and used them to grow aggressively. The same rate actions restored the capital position of the competitors it was taking share from, and those competitors are now able to compete on price again. Premium per policy falling while policies grow is the visible signature of that.

The structural advantages remain intact: a genuine direct-to-consumer channel at scale, the deepest telematics dataset in the industry, and a cost position that supports a 96 target combined ratio while others run closer to breakeven. Nothing in the current data suggests those have eroded. What has changed is that the excess return available from them is smaller when everyone else can afford to price rationally.

At 11.0 times trailing earnings and 3.71 times book, with a 34.9% return on equity, Progressive is neither expensive nor obviously cheap, and the two multiples say different things.

The earnings multiple looks undemanding until you accept that the earnings are cyclical. On the consensus 2026 figure of $17.97, the forward multiple is about 12.2 times, and consensus has that number falling. The price to book of 3.71 is the more informative number: it is a premium multiple that can only be justified by the return on equity persisting, and the return on equity is a function of the same underwriting margin that is now normalising. A 96 combined ratio produces a materially lower return on equity than 87.3 does.

The bull case does not require heroics. It requires the 2026 vintage of policies to retain and to run at target loss ratios, which would mean the company has bought durable share at the bottom of the cycle and the current margin compression is an investment rather than an erosion. The bear case requires only that competitors keep pricing rationally and that the 96 target is reached from above, with margin compressing into it.

We do not publish price targets. At this quotation the distribution of outcomes looks roughly symmetric to us, which is the definition of the Neutral stance, and it is why we would rather hold the work and revisit it after two or three more quarters of retention data than manufacture a view the evidence does not support.

Open workstreams, ordered by how much they move the view.

1. 01

   Read the full 10-K risk factors on autonomy. We reviewed a filing summary, not the complete document. Whether Progressive names autonomous vehicles as a risk, and how it frames the exposure, is the single cheapest piece of work available and it bears directly on Section 08.

2. 02

   Track the 2026 vintage. Retention and loss ratios on policies written this year decide whether the margin compression is investment or erosion. This is the question that resolves the Neutral stance in either direction.

3. 03

   Get the commercial book by sub-line. The 13% commercial share needs decomposing into for-hire trucking, business auto, contractors and transportation network drivers. Only the trucking piece is exposed to freight autonomy, and we have not sized it.

4. 04

   Size the personal auto pool erosion properly. Build the arithmetic that Section 08 sketches: fleet turnover rate, ADAS penetration by model year, frequency and severity by vehicle generation, against premium per policy. This is the KDK pool-shift model applied to personal lines.

5. 05

   Compare against GEICO's recovery. Berkshire's auto operation is the competitor whose repaired balance sheet most directly explains Progressive's pricing environment. Its expense ratio and policy growth are the tell.

6. 06

   Revise the Kodiak allocation map. Section 07 amends a published house view. That amendment should be carried into the Kodiak note as a dated revision, so it does not stand only here.

• Progressive reports June 2026 results, investor relations
• Progressive Q2 2026 results and 10-Q
• Progressive 2025 Form 10-K, segment disclosure
• PGR statistics, valuation and short interest
• PGR analyst forecasts and consensus target
• Progressive Snapshot telematics deployment
• Progressive Commercial fleet size and appetite
• IIHS and HLDI, sensor-equipped vehicle claim frequency and severity
• Contra Research Group, Kodiak AI initiation, 5 September 2026
• FINRA equity short interest reporting