Hapoalim: Why the Collective Reserve Keeps Rising Even as Credit Metrics Stay Strong

The main article dealt with the broad 2025 picture. This follow-up isolates the point that the headline numbers partly blur: how can nonaccrual credit fall to 0.48%, problematic credit fall to 1.48%, and nonaccrual coverage rise to 310%, while the credit-loss expense rate still climbs to 0.27% and the collective reserve keeps building.

The short answer is that Hapoalim is not reserving against a blowup that has already happened. It is reserving against risk that still sits ahead of the book. That distinction matters. Nonaccruals and problematic-credit ratios are backward-looking credit-quality indicators. The collective reserve is management’s reading of what the portfolio could still absorb. So the gap between the two stories is not an accounting contradiction. It is a risk statement: the bank thinks the book still looks strong, but not strong enough to justify a softer stance.

That argument rests on more than a vague reference to “uncertainty.” Three sharper findings drive it.

Finding one: the improvement in asset-quality metrics is real, but it did not lead to reserve release. Problematic credit fell 9.0% to NIS 8.419 billion, nonaccrual credit fell 10.7% to NIS 2.837 billion, yet the credit-loss expense rate rose from 0.16% to 0.27%.

Finding two: almost the entire movement in the allowance balance came from the collective side. The total allowance rose by NIS 904 million to NIS 8.840 billion, with the collective allowance up NIS 964 million while the individual allowance fell by NIS 60 million. That is not a reaction to a wave of borrower-specific failures. It is a decision to keep thickening the buffer on a growing book.

Finding three: the main pocket behind that caution is construction and real estate. Total exposure to construction and real estate in Israel rose 12.5% to NIS 189.241 billion, roughly 28.1% of total public credit risk in Israel. About 93.4% of that sits in the business division, and roughly 53% of the division’s real-estate exposure is residential construction.

That chart is the core of the continuation thesis. On first read, the easy reaction is to focus on the left side of the picture: nonaccruals are lower, problematic credit is lower, and even the total allowance ratio to credit edged down from 1.76% to 1.72%. But that is exactly the wrong shortcut. What fell was the ratio, not the bank’s margin of safety. When the book grows quickly, the ratio can drift lower even while the bank is still adding buffer. In practice, coverage of nonaccrual credit jumped from 254.7% to 310.4%. That is not the profile of a bank that feels ready to release reserves.

The Collective Reserve Is Telling A Different Story From The Classified-Loan List

To understand that gap, the first step is to separate two very different mechanisms. The individual allowance reacts to specific borrowers that have already run into trouble. The collective allowance is built on an expected-loss model, on the whole portfolio, and on forward scenarios. In 2025, the first mechanism actually moved in Hapoalim’s favor. The investor presentation says explicitly that the individual allowance recorded income because of recoveries from a small number of borrowers. By contrast, the full-year and fourth-quarter provision were driven by a higher collective allowance, mainly because of loan-book growth and continued uncertainty in the economic environment.

That distinction matters because it explains why the problem is not sitting primarily in the list of borrowers that have already cracked. If Hapoalim were mainly reacting to borrower-specific deterioration, the individual allowance should have been doing more of the work. That is not what happened. The bank chose to expand a broad reserve against a book that still looks clean in rear-view metrics, but looks less comfortable when read forward.

The composition of the allowance supports that reading. At the end of 2025, the total allowance stood at NIS 8.840 billion, of which NIS 7.765 billion sat on a collective basis, about 87.8% of the full cushion. A year earlier, the collective share was about 86.4%. In plain English, the reserve did not just get bigger. It also became more collective.

There is one more detail worth stopping on. In the annual report, the bank says the collective allowance is built using three macro scenarios, optimistic, base, and pessimistic. At the end of 2025, the collective allowance balance was about NIS 8.5 billion, which management says still reflects high uncertainty despite the more positive geopolitical developments in the fourth quarter. More than that, if the Israeli portfolio had been reserved only under the pessimistic scenario, the collective allowance would have been about NIS 1.2 billion higher than the amount actually recorded. If it had been calculated only under the optimistic scenario, the allowance would have been about NIS 0.9 billion lower.

That means the current reserve is not only a function of what has already broken. It also captures how much management still thinks the environment can worsen. And when the bank explicitly says its pessimistic scenario is not an extreme-stress scenario, it is effectively telling readers that this remains a buffer for uncertainty management, not a balance sheet that has already normalized.

Real Estate Is Not Background Noise. It Is The Center Of Gravity Behind The Caution

That is where the construction and real-estate portfolio comes in. On the surface, even this part of the book still looks manageable. Total nonaccrual credit in construction and real estate fell 35.1% to NIS 377 million, and the problematic-credit ratio in the sector fell from 1.32% to 1.16%. If the analysis stops there, it is easy to conclude that sector risk is simply getting better.

But the risk report describes a backdrop that does not justify much comfort. New-home sales fell about 25% in 2025, unsold inventory climbed to roughly 83,000 units in December, apartment prices rose just 0.4% for the year and still sat 1.4% below the February 2025 peak, and the bank explicitly ties sector risk to high interest rates, construction-cost pressure, and weaker household purchasing power.

The implication is that Hapoalim is not looking only at defaults that have already materialized. It is looking at a market in which unsold inventory has become larger, financing structures have become more sensitive, and the regulator itself has intervened in residential project economics. Beginning April 6, 2025, Proper Conduct of Banking Business Directive 203 requires higher capital allocation for residential projects if the share of contracts with non-linear payments, meaning more than 40% of the apartment price deferred to delivery, crosses the regulatory threshold. The bank explicitly says it adjusted credit policy for projects with such non-linear collections. That is not the language of a bank reading only past-due data.

That chart sharpens why Hapoalim keeps carrying such a thick cushion even while classic asset-quality ratios look clean. Real estate is not just large. It is still expanding. Total exposure reached NIS 189.241 billion, and the largest single component is residential construction at NIS 92.861 billion. Income-producing assets add another NIS 42.927 billion, and the business division carries almost the entire center of gravity. This is not a corner of the book. It is one of its defining axes.

It is equally important to understand what the bank is not saying. Hapoalim is not flagging a sector collapse. Quite the opposite, it shows lower problematic and lower nonaccrual credit. But it is clearly flagging a sector mix that still makes caution rational: higher unsold inventory, weaker new-home sales, contractor promotions that the regulator has chosen to penalize in capital terms, and a loan book that has already grown deeper into that same sector.

The Real Gap Sits In The Watchlist Layer And In Project Structure

If the goal is to understand why the headline quality metrics are not telling the full story, the most important places to look are the ones that rarely make the headline slide. The first is credit that is not in performance rating. The second is the structure of project exposure inside the real-estate portfolio.

Within construction and real estate, total credit not in performance rating rose 10.6% to NIS 3.394 billion. Inside that, the portion that is not problematic but still not in performance rating jumped 41.5% to NIS 1.190 billion. In other words, while visible problematic credit fell, the early-warning bucket widened materially. That does not contradict the improvement in headline metrics. It explains why the bank is unwilling to read those metrics as a clean green light.

The second place is the business-division real-estate book broken down by loan-to-value and absorption capacity. That is where the risk posture becomes easier to understand. In the land-financing book, NIS 20.693 billion, about two thirds of the balance, sits above 65% LTV. In projects under construction, NIS 33.704 billion, or 41.3% of exposure, sits in projects with absorption capacity of up to 50%, versus 28.9% a year earlier. The weakest bucket, absorption capacity of up to 25%, more than doubled to NIS 2.174 billion.

One caveat matters here, and the report states it directly. The reported LTV and absorption-capacity figures do not net the effect of credit-risk mitigation from purchased insurance or from credit-linked notes. So they do not show inevitable loss. What they do show is how much gross risk the bank still believes deserves close management.

This is exactly the kind of data point that bridges the gap between the clean headline ratios and the reserve behavior. If classified loans alone were telling the whole story, readers should have expected reserve release, or at least a sharp halt in further building. Instead, what the bank is effectively saying is simpler: visible borrower stress has not spread yet, but parts of the project book look more sensitive, so it prefers to carry the cushion before the problem becomes obvious.

The presentation all but says that directly. It highlights that credit-quality metrics are strong while the allowance remains more than three times the nonaccrual balance. That is not the profile of a bank trying to maximize near-term earnings. It is the profile of a bank that would rather preserve a surplus of caution as long as the main sector-risk pocket, especially real estate, has not truly normalized.

Conclusion

The bottom line is that Hapoalim’s collective reserve is not arguing with the strong credit metrics. It is arguing with the comfortable interpretation of those metrics. Yes, nonaccrual credit is lower. Yes, problematic credit is lower. But the loan book also grew 13.4% to NIS 502.881 billion, construction and real-estate concentration remains very large, and the early-warning layers inside the sector are still widening.

So the cleaner way to read 2025 is this: Hapoalim is not maintaining a high cushion because it already sees broad impairment across the book. It is maintaining a high cushion because it sees a strong book that still enters 2026 with enough concentration and enough uncertainty to justify caution over complacency. That matters especially now, because as the rate tailwind fades, earnings quality will be judged more through reserve quality and less through the absolute size of the bottom line.

There is a real counter-thesis. If housing demand stabilizes, unsold inventory starts to clear, contractor promotions lose weight, and the larger borrowers keep performing, part of today’s cushion could end up looking like excess conservatism. But as of the end of 2025, Hapoalim’s filings are pointing in a clearer direction: credit quality still looks strong, and the bank is still choosing to manage it as if this is not yet the moment to lower the shield.