Cleansing or Destruction? Decomposing Germany’s 21-Year Insolvency High
Executive Summary
In the second quarter of 2026 the Leibniz Institute for Economic Research Halle (IWH) counted 4,996 insolvencies of partnerships and corporations in Germany, the highest quarterly figure since the second quarter of 2005. The official court statistics of the Federal Statistical Office tell a similar story at a lower amplitude: 12,812 corporate insolvency applications in the first half of 2026, the most for a first half-year since 2013, and an estimated 25,800 for the year as a whole, 7 percent more than in 2025 and 29 percent more than the 2016 to 2019 average. The headline has been read in two opposite ways. For one camp, it is the long-delayed cleansing of an economy in which a decade of zero interest rates and two years of pandemic support kept unviable firms alive. For the other, it is the destruction of viable firms by an energy shock, a cost shock and a demand slump that have nothing to do with their underlying quality.
This paper argues that both readings are partly right, but not in equal measure, and that the proportions can be measured. We build a 27-year annual series of corporate insolvencies and a monthly series from 2021 to October 2026, and estimate a parsimonious insolvency equation on data from 2001 to 2019 only, before the pandemic, the energy crisis and the rate reversal. The equation relates insolvencies to their own past, to real GDP growth and to the lagged ten-year Bund yield. Simulated forward with actual macroeconomic data, it serves as a counterfactual: the number of firms that should have failed each year given the economy they faced.
Three results follow. First, the equation reveals a large insolvency deficit during the pandemic: about 8,200 firms that should have failed between 2020 and 2022 did not. Second, the same equation, with no information from after 2019, tracks the steep rise from 2023 to 2025 almost exactly; its 2025 prediction is within 1 percent of the outturn once a modest catch-up is allowed for. A competing specification that extrapolates the pre-2020 downward trend in insolvencies fails this post-sample test by almost five standard deviations. The data thus point clearly to a mechanism: the wave is above all the lagged response of German firms to the end of zero interest rates.
Third, we decompose the 2026 total into five components: a zero-rate baseline of about 17,100 cases, rate normalisation (about 5,400), weak demand (about 1,000), the cost shock triggered by the Iran war (about 1,500, identified from the acceleration after February 2026), and the release of pandemic catch-up (about 800). Classifying rate normalisation and catch-up as cleansing, and the demand and war shocks as destruction, 71 percent of the excess is cleansing. A Monte Carlo simulation with 19,980 draws over parameter uncertainty and the 2026 outturn gives a median of 70 percent and an 84 percent probability that cleansing dominates.
That conclusion needs three qualifications, and they are what turn a reassuring headline into a policy agenda. The first is that the cleansing is incomplete: about 4,200 of the 8,200 missing pandemic insolvencies have still not occurred, so a stock of weak firms remains. The second is that the destruction, although smaller, falls disproportionately on the industrial core. In October 2026, about 9,400 manufacturing jobs were affected by insolvencies, the second-highest monthly value since the IWH series began, and the reported success rate of restructurings of larger insolvent firms has fallen from 57 percent in 2020 to 32 percent in 2025. The third is that reallocation, the payoff of cleansing, is weak: the firm stock implied by Destatis insolvency rates rose by only about 1.7 percent between 2025 and 2026, while unemployment has reached about 3.0 million.
The composition of the wave reinforces this reading. Incorporated firms are failing at rates 70 to 80 percent above their pre-pandemic norm, while the residual category of sole proprietors and other small units appears to be about 22 percent below it. The average real claim per insolvency has fallen to EUR 1.4 million in 2026, 18 percent below its 2016 to 2019 average and half its 2024 level. Germany is losing many small and medium-sized incorporated firms rather than a few large ones, and these are precisely the firms whose survival, between 2010 and 2021, depended most on cheap credit.
The policy implications are counter-intuitive. A broad rescue programme would mostly protect the firms that the rate normalisation is correctly weeding out, and would rebuild the overhang this paper measures. The right response is to accept the cleansing and to make the destruction cheaper: a faster and more accessible pre-insolvency restructuring framework, temporary and narrowly targeted relief for energy-intensive firms with viable business models, and active policies to move workers and capital from failing firms to growing ones. For 2027, our baseline projects about 26,700 corporate insolvencies, because the rate increases of 2026 are already locked into next year’s financing costs. The wave is not over, but it is, for the most part, the kind of wave a market economy needs.
Key findings.
- Seven tenths of the 2026 insolvency wave is a return to normal, not a collapse. Of the roughly 8,650 corporate insolvencies in 2026 that would not have occurred had the zero-rate world continued, our central model attributes 71 percent (about 6,160 cases) to the normalisation of interest rates and to the delayed exit of firms kept alive during the pandemic. The Monte Carlo median is 70 percent, with an 80 percent band of 42 to 86 percent.
- Interest rates, not energy, are the dominant driver. An insolvency equation estimated on 2001 to 2019 data, before any of the recent shocks, predicts the 2023 to 2026 surge with an error of less than 1 percent in 2025. Its long-run semi-elasticity implies that each percentage point on the ten-year Bund yield raises insolvencies by about 19 percent. Rate normalisation alone accounts for about 5,400 cases in 2026.
- Half of the pandemic insolvency deficit has still not materialised. Between 2020 and 2022 Germany recorded about 8,200 fewer corporate insolvencies than its fundamentals implied. Only about 4,000 of these missing cases had been absorbed by the end of 2026, leaving an overhang of roughly 4,200 firms.
- The Iran war added about 1,500 insolvencies in ten months. Year-on-year insolvency growth accelerated by 7 to 8 percentage points after the outbreak of the war on 28 February 2026, in both the official court statistics and the IWH early series. This is the clearest identifiable piece of genuine destruction.
- The 21-year high is a wave of incorporated firms, while micro-business insolvencies are below their pre-pandemic level. Insolvencies of partnerships and corporations are running about 70 to 80 percent above their 2016 to 2019 average; the implied residual of sole proprietors and other small units is about 22 percent below it. The average real creditor claim per case has fallen to EUR 1.4 million, 18 percent below the pre-pandemic norm and half the 2024 level.
- 2027 will probably be higher still. Because the 2026 rise in Bund yields is already locked into next year's financing costs, our baseline projects about 26,700 corporate insolvencies in 2027 (80 percent band 25,100 to 28,400), with a 76 percent probability of exceeding 2026. A prolonged energy shock would push the figure to about 28,700.
1. Introduction: Two Stories About the Same Number
Every insolvency wave invites two stories. In the first, insolvencies are the immune system of a market economy. Firms that no longer cover their cost of capital exit, their workers and assets move to more productive uses, and aggregate productivity rises. Joseph Schumpeter called this creative destruction; the literature on “zombie firms” since the 2010s has given it a modern empirical form, documenting how persistently low interest rates and forbearance by banks allow unprofitable firms to survive, depress investment and wages at healthy competitors, and slow productivity growth. In this story an insolvency wave after a long period of cheap money is not a disease but a delayed cure.
In the second story, insolvencies are a transmission channel of shocks. When energy prices spike, a war closes a shipping lane or demand collapses, even well-managed firms with sound business models can run out of liquidity. Their failure destroys firm-specific knowledge, customer relationships and organisational capital that cannot be rebuilt quickly. Workers lose firm-specific skills, suppliers lose customers, and the economy’s capacity is permanently reduced. In this story an insolvency wave during an energy and cost shock is a loss, and policy should prevent it.
Germany in 2026 has the ingredients of both stories. On the cleansing side, the country went through a decade in which the ten-year Bund yield averaged close to zero, followed by two years of extraordinary pandemic support and a temporary suspension of the obligation to file for insolvency. Corporate insolvencies fell to 13,993 in 2021, the lowest since the Insolvency Code came into force in 1999. On the destruction side, the US-Israel war on Iran since 28 February 2026 has disrupted the Strait of Hormuz, Qatari LNG deliveries have been under force majeure for months, producer prices of energy were 14.9 percent higher in September 2026 than a year earlier, and mineral oil products 51.3 percent higher. The European Central Bank raised its deposit rate in June and September 2026, to 2.5 percent, and the ten-year Bund yield reached about 3 percent, its highest level since 2011. The statutory minimum wage rose by about 8.5 percent in January 2026. GDP is growing, but only by 0.3 percent in the second quarter and 0.2 percent in the third, and largely because of debt-financed public spending.
Which story dominates matters for policy. If the wave is mostly cleansing, the right response is to let it run, while easing the reallocation of workers and capital. If it is mostly destruction, the right response is targeted support to bridge a temporary shock. Treating a cleansing wave as destruction would rebuild the zombie stock; treating a destruction wave as cleansing would waste viable capacity at a time when Germany can least afford it.
This paper measures the proportions. Section 2 sets out the facts: a 27-year series, the monthly profile of 2026, and the important difference between the two main data sources. Section 3 estimates an insolvency equation on pre-pandemic data and tests competing specifications. Section 4 uses the equation to measure the pandemic insolvency deficit and its catch-up. Section 5 identifies the war shock from the timing of the 2026 acceleration. Section 6 combines these elements in a decomposition of 2026 insolvencies and a Monte Carlo assessment of the cleansing share. Section 7 asks whether the wave is reallocating resources productively. Section 8 presents scenarios for 2027. Section 9 draws policy implications and Section 10 concludes.
2. Anatomy of the Wave
2.1 A 27-year perspective
The long series puts the 2026 headlines in perspective. Corporate insolvency applications recorded by German local courts rose from 26,476 in 1999 to a peak of 39,320 in 2003, as the bursting of the technology bubble, a stagnating economy and a wave of subsidised start-ups combined. They remained above 36,000 until 2005, then fell rapidly in the 2006 to 2007 upswing, rose again to 32,687 in the financial crisis year of 2009, and then declined almost without interruption for a decade, to 18,749 in 2019. The pandemic pushed the series to its low: 15,841 in 2020, 13,993 in 2021 and 14,590 in 2022. Then the trend reversed sharply: 17,814 in 2023, 21,812 in 2024, 24,064 in 2025 and an estimated 25,800 in 2026.
Table 1. Corporate insolvencies in Germany, selected years.
| Year | Corporate insolvencies | Change on previous year (%) | Comment |
|---|---|---|---|
| 2003 | 39,320 | +4.6 | Series peak |
| 2005 | 36,843 | minus 6.0 | Last year above 35,000 |
| 2009 | 32,687 | +11.6 | Financial crisis |
| 2019 | 18,749 | minus 2.9 | Pre-pandemic low |
| 2021 | 13,993 | minus 11.7 | Series low, filing suspension |
| 2023 | 17,814 | +22.1 | Reversal |
| 2024 | 21,812 | +22.4 | Steepest two-year rise in the series |
| 2025 | 24,064 | +10.3 | Highest since 2014 |
| 2026 (estimate) | about 25,800 | +7.1 | Range 25,400 to 26,200 |
Source: Destatis; 2026: author's estimate from January to July data and IWH monthly trends.
Two points stand out. The first is that, measured by the official count, 2026 is not a 21-year high. The 2026 total is about one third below the 2003 to 2005 plateau and well below the 2009 level. The second is that the speed of the reversal is unprecedented: insolvencies rose by 65 percent between 2022 and 2025, compared with 49 percent between 1999 and 2003. The rise is steep because it starts from an artificially low base, a point to which Section 4 returns.
2.2 The 2026 profile
Table 2 shows the monthly data for 2026 from both sources. The official statistics count insolvency applications decided by local courts and enter the statistics about three months after the application. The IWH Insolvenztrend counts insolvencies of partnerships and corporations from court announcements and is available about five weeks after the end of each month.
Table 2. Monthly corporate insolvencies in 2026.
| Month 2026 | Destatis applications | y/y (%) | IWH cases | y/y (%) | IWH jobs affected |
|---|---|---|---|---|---|
| January | 1,919 | +4.9 | 1,391 | +4 | 16,865 |
| February | 2,048 | minus 1.0 | 1,466 | +2 | 23,290 |
| March | 2,308 | +15.8 | 1,716 | +18 | 13,922 |
| April | 2,276 | +7.1 | 1,776 | +10 | 19,966 |
| May | 1,995 | minus 2.0 | 1,518 | +3 | 11,309 |
| June | 2,266 | +15.8 | 1,702 | +20 | 14,230 |
| July | 2,373 | +8.0 | 1,689 | +7 | 13,313 |
| August | not yet published | 1,525 | +9 | 16,146 | |
| September | not yet published | 1,666 | +13 | 11,569 | |
| October | not yet published | 1,683 | +9 | about 16,800 |
Source: Destatis, Insolvenzen nach Monaten (as of 9 October 2026); IWH Insolvenztrend releases, as reported. IWH job counts refer to the largest 10 percent of insolvent firms.
The quarterly sums are 6,275 applications in the first quarter and 6,537 in the second in the official data, and 4,573, 4,996 and 4,880 in the first three quarters in the IWH series. January to July 2026 brought 15,185 applications, 6.9 percent more than a year earlier. The IWH reported that October was 9 percent above October 2025 and 83 percent above the October average of 2016 to 2019, and that its leading indicators point to fewer insolvencies in November and December.
To estimate the 2026 total, we apply the IWH year-on-year growth rates for August to October to the corresponding 2025 official figures, and a growth rate of 3 percent for November and December in line with the IWH’s expectation of a slowdown. The result is about 25,800 cases, with a range of 25,400 to 26,200 when the monthly growth rates are varied by 4 percentage points either way. The twelve-month rolling sum to July 2026 already stood at 25,043.
2.3 Creditor claims, sectors and consumers
Three further features of the wave matter for its interpretation. First, the expected creditor claims have fallen sharply even as case numbers have risen. They totalled about EUR 18.5 billion in the first half of 2026, against EUR 28.2 billion a year earlier, and EUR 21.9 billion from January to July against EUR 31.8 billion. Destatis attributes the decline to fewer economically significant firms filing. For the whole of 2025, claims were about EUR 47.9 billion, down from EUR 58.1 billion in 2024.
Second, the wave is concentrated in a small number of sectors. In the first seven months of 2026 there were 42.8 insolvencies per 10,000 firms across the economy, but 85.5 in transport and storage, 70.0 in hospitality and 62.9 in construction. In the IWH data, the second quarter of 2026 brought new highs since 2020 in almost all major sectors, including construction, real estate, trade, hospitality and services, with manufacturing the main exception because its peak had come in 2025. The IWH reported record numbers in business-related services and in financial and insurance services in July.
Third, consumer insolvencies are rising more slowly than corporate ones. They reached 77,219 in 2025, 8.4 percent more than in 2024, but only 46,043 from January to July 2026, an increase of 1.0 percent. Household distress is therefore not the main driver of the 2026 dynamics, which are concentrated in the corporate sector.
2.4 Two sources, two waves
The most informative fact about the 2026 wave is hidden in the difference between the two data sources. The IWH series, which covers only partnerships and corporations, is running about 70 to 80 percent above its 2016 to 2019 average: the IWH reports excesses of 71 percent for March, 80 percent for June, 75 percent for July, 85 percent for September and 83 percent for October. The official count, which also includes sole proprietors, is only about 29 percent above its 2016 to 2019 average.
We can use this difference to make an approximate split. From the five months for which the IWH publishes the pre-pandemic comparison, the implied 2016 to 2019 base of the IWH series is about 947 cases a month, or 11,360 a year. Since the official count averaged 19,916 in 2016 to 2019, the residual category, mostly sole proprietors and other small units not covered by the IWH, accounted for about 8,550 cases a year. For 2026, the IWH series will total about 19,100 cases if November and December are 8 percent below the year’s average. The official estimate of 25,800 then leaves a residual of about 6,700, roughly 22 percent below its pre-pandemic level.
Table 3. Composition of the 2026 wave by legal form.
| Measure | 2016 to 2019 average | 2026 (estimate) | Change (%) |
|---|---|---|---|
| Official count, all corporate insolvencies (Destatis) | 19,916 | 25,778 | +29 |
| Partnerships and corporations (IWH) | about 11,360 | about 19,100 | +68 |
| IWH, average of five months with published comparison | +79 | ||
| Residual: sole proprietors and other units (implied) | about 8,550 | about 6,680 | minus 22 |
Source: Destatis; IWH; author's calculations. The two sources differ in coverage and timing, so the residual is indicative.
The calculation is approximate, since the two series differ in their counting rules and in their timing. But the gap is too large to be an artefact. The 2026 wave is a wave of incorporated firms, while insolvencies of the smallest business units are, if anything, below their pre-pandemic level. This is the first clue to the nature of the wave: it is concentrated in the segment of firms that borrow from banks and markets, and therefore in the segment most exposed to the interest rate cycle.
3. An Insolvency Equation Estimated Before the Shocks
3.1 Specification
To separate the different drivers, we need a benchmark: how many insolvencies would Germany have had if the economy had behaved after 2019 as it did before? We estimate an annual insolvency equation on data from 2001 to 2019, a period that contains a full cycle of booms, the financial crisis and the long decline of interest rates, but none of the shocks of the 2020s. The baseline specification, which we call Model R, is:
log I(t) = a + rho x log I(t minus 1) + b x g(t) + c x i(t minus 1)
where I is the number of corporate insolvencies, g is real GDP growth in percent and i is the annual average yield on ten-year Bunds in the previous year. The lagged dependent variable captures the persistence of insolvencies, which reflects the gradual depletion of firms’ equity buffers and the length of court procedures. The lagged yield captures the cost of refinancing, which affects firms with a delay as loans mature.
We tested two alternatives. Model T adds a linear time trend, which absorbs any secular decline in insolvencies unrelated to the business cycle, for example the falling number of start-ups. Model C adds two cost variables: the profit squeeze, defined as unit labour cost growth minus the GDP deflator, and the change in the real producer price of energy.
3.2 Results
Table 4. Insolvency equations, annual data 2001 to 2019 (19 observations). Dependent variable: log of corporate insolvencies. t-statistics in parentheses.
| Coefficient | Model R | Model T | Model C |
|---|---|---|---|
| Lagged log insolvencies (rho) | 0.683 (7.5) | 0.643 (7.3) | 0.657 (6.5) |
| Real GDP growth, pp | minus 0.0105 (minus 1.9) | minus 0.0131 (minus 2.4) | minus 0.0172 (minus 1.4) |
| Ten-year Bund yield, lagged, pp | 0.0555 (4.6) | 0.0158 (0.6) | 0.0543 (4.0) |
| Time trend, years | minus 0.0135 (minus 1.8) | ||
| Profit squeeze (ULC growth minus deflator), pp | minus 0.0114 (minus 0.8) | ||
| Real energy price change, % | minus 0.0004 (minus 0.2) | ||
| R squared | 0.970 | 0.975 | 0.971 |
| Standard error of regression | 0.046 | 0.043 | 0.048 |
| Durbin-Watson | 1.48 | 1.53 | 1.75 |
| Long-run semi-elasticity to the Bund yield | 0.175 | 0.044 | 0.158 |
Source: author's estimates from Destatis, OECD and FRED data.
Model R explains 97 percent of the variation in log insolvencies over two decades with three variables. The interest rate coefficient is large and precisely estimated: in the long run, a one percentage point higher Bund yield raises insolvencies by about 17.5 log points, or 19 percent. GDP growth has the expected negative sign: each percentage point of growth lowers insolvencies by about 1 percent in the same year and 3.3 percent in the long run.
The cost variables in Model C are insignificant and have the wrong sign. This should not be read as evidence that energy and labour costs do not matter. Between 2001 and 2019 Germany experienced no energy price shock comparable to 2022 or 2026, and unit labour costs moved mostly in the opposite direction to profits only during the 2009 recession, when labour hoarding through short-time work raised unit labour costs while insolvencies rose for other reasons. The pre-2020 sample simply cannot identify these effects. We therefore keep the cost shocks out of the equation and identify the 2026 cost shock separately, from the timing of the acceleration after the outbreak of the Iran war (Section 5).
3.3 Which model? Out-of-sample and post-sample tests
Models R and T fit the estimation period almost equally well, but they tell very different stories about the 2020s. In Model R, the decline of insolvencies from 2003 to 2019 is explained mostly by falling interest rates, and their rise after 2022 by the return of positive rates. In Model T, the same decline is attributed largely to a secular trend, interest rates matter little, and the post-2022 surge becomes a large unexplained excess.
We test the two models in two ways. In the first, a conventional out-of-sample test, we estimate each model on a shorter sample and forecast the remaining pre-2020 years dynamically. In the second, a post-sample test, we compare the predictions of the full 2001 to 2019 models with the 2023 to 2025 outturn, allowing for a release of pandemic catch-up equal to one quarter of each model’s estimated deficit in each year from 2023.
Table 5. Model tests. Errors are log differences between actual and predicted values.
| Test | Model R | Model T |
|---|---|---|
| Out-of-sample RMSE, estimated to 2012, forecast 2013 to 2019 (%) | 16.3 | 15.2 |
| Out-of-sample RMSE, estimated to 2014, forecast 2015 to 2019 (%) | 10.2 | 2.9 |
| Out-of-sample RMSE, estimated to 2016, forecast 2017 to 2019 (%) | 2.4 | 0.3 |
| Post-sample error 2023, after catch-up allowance (%) | minus 12.6 | minus 4.3 |
| Post-sample error 2024, after catch-up allowance (%) | minus 3.4 | +15.8 |
| Post-sample error 2025, after catch-up allowance (%) | minus 0.5 | +27.1 |
| Dynamic forecast standard deviation, sixth year (%) | 6.3 | 5.6 |
| Relative likelihood weight from 2025 post-sample error | 1.00 | below 0.001 |
Source: author's calculations.
The two tests point in opposite directions, and the reason is instructive. Within the pre-2020 period, interest rates and the time trend move together, so Model T forecasts the years 2015 to 2019 better: the trend captures the steady decline more smoothly than the yield. After 2022, however, interest rates and the trend part company. Rates rose by almost 3 percentage points while the trend kept pointing down. Model T predicts flat insolvencies of about 16,500 a year from 2023 to 2025 and misses the 2025 outturn by 27 percent even after generous allowance for catch-up, an error almost five times its own forecast standard deviation. Model R, which knows nothing about the period after 2019, predicts 22,131 insolvencies in 2025 against an actual 24,064, and the gap is closed by the catch-up allowance.
The post-sample evidence is decisive because the rate reversal of 2022 to 2025 is exactly the natural experiment that separates the two hypotheses. We therefore use Model R as our central model. We retain Model T as a sensitivity case, because it represents a coherent alternative view: that the pre-2020 decline was partly structural, and that the post-2022 wave is therefore larger than fundamentals justify.
4. The Pandemic Insolvency Deficit and Its Catch-Up
4.1 Measuring the missing insolvencies
Simulated forward from the 2019 outturn with actual GDP growth and Bund yields, Model R predicts 18,862 corporate insolvencies in 2020, 17,169 in 2021 and 16,586 in 2022. The fundamentals of those years, a deep recession in 2020 followed by a partial recovery with still negative interest rates, would have produced a moderate rise in 2020 and a decline thereafter. Actual insolvencies were far lower: 15,841, 13,993 and 14,590. The difference, about 8,200 firms over three years, is the pandemic insolvency deficit.
The reasons are well known. The obligation to file for insolvency was suspended for firms whose distress was caused by the pandemic for much of 2020 and, for firms awaiting support payments, into 2021. Short-time work schemes absorbed labour costs, and the federal and state governments paid emergency aid, bridging aid and loan guarantees on a scale without precedent. Banks extended forbearance. All of these measures were justified as a response to a shock that was temporary by design. But they could not distinguish between firms that were viable before the pandemic and firms that were already failing. They therefore suspended the market’s selection process for three years.
4.2 The catch-up ledger
If the deficit consists of firms that should have failed, it should eventually be released. We keep a simple ledger. Each year from 2023 onwards, any excess of actual insolvencies over the model prediction is counted as catch-up until the deficit is exhausted. In 2026, the part of the excess that coincides with the war shock identified in Section 5 is attributed to the war first, so that the catch-up estimate is conservative.
Table 6. The catch-up ledger, Model R.
| Year | Model R prediction | Actual | Actual minus prediction | Cumulative balance | Of which counted as catch-up |
|---|---|---|---|---|---|
| 2020 | 18,862 | 15,841 | minus 3,021 | minus 3,021 | |
| 2021 | 17,169 | 13,993 | minus 3,176 | minus 6,197 | |
| 2022 | 16,586 | 14,590 | minus 1,996 | minus 8,193 | |
| 2023 | 18,156 | 17,814 | minus 342 | minus 8,535 | 0 |
| 2024 | 20,525 | 21,812 | +1,287 | minus 7,248 | 1,287 |
| 2025 | 22,131 | 24,064 | +1,933 | minus 5,315 | 1,933 |
| 2026 (estimate) | 23,481 | 25,778 | +2,297 | minus 3,018 | 766 (after war shock) |
| Remaining deficit at end-2026 | about 4,200 |
Source: author's calculations from Destatis data.
The ledger yields the third key finding of this paper. Of the 8,200 missing insolvencies of 2020 to 2022, only about 4,000 had been released by the end of 2026. The deficit even widened slightly in 2023, when insolvencies were still below what fundamentals implied, and the release began only in 2024. At the end of 2026 about 4,200 firms remain in the overhang.
This number should be read carefully. Not every missing insolvency will eventually show up in the statistics. Some firms that survived thanks to support may have genuinely recovered; others may have closed without formal insolvency, through voluntary liquidation or simple deregistration, which the insolvency statistics do not record. The overhang is therefore an upper bound on the future catch-up, not a forecast. But it is an indication that the cleansing of the pandemic period is not complete, and that weak firms continue to occupy resources.
4.3 Is the rise in 2023 to 2026 just catch-up?
A popular reading of the German wave holds that it is nothing more than the catch-up of the pandemic. The ledger rejects this reading. Over 2023 to 2026, the total excess of actual insolvencies over Model R is about 5,200 cases, far smaller than the rise in insolvencies itself: from 14,590 in 2022 to about 25,800 in 2026, an increase of about 11,200. Most of the rise is explained by the model, that is, by the change in fundamentals, above all in interest rates. Catch-up contributes to the level, but it is not what drives the trend.
The alternative Model T gives a very different answer. Because it predicts flat insolvencies, it attributes an excess of about 24,200 cases to 2023 to 2026, almost three times its estimated deficit. In that world, 2026 would contain about 8,400 insolvencies that neither fundamentals nor catch-up can explain, which would have to be read as genuine destruction. Section 3 explained why the data reject Model T. But the contrast shows how much the answer to the cleansing question depends on the role one assigns to interest rates.
5. Identifying the War Shock
5.1 An event study on the 2026 acceleration
The insolvency equation does not contain energy or labour cost variables, because the pre-2020 sample cannot identify them. To measure the contribution of the 2026 cost shock, we exploit its timing. The US-Israel war on Iran began on 28 February 2026. Energy and transport costs rose sharply from March onwards: by September, producer prices of energy were 14.9 percent higher than a year earlier, natural gas for distribution 15.0 percent and mineral oil products 51.3 percent, after year-on-year increases of 8.3 percent for energy and 7.5 percent for natural gas in August. If the war was a material driver of insolvencies, year-on-year growth should have accelerated after February.
Table 7. Event study of the war shock.
| Period | Destatis, y/y (%) | IWH, average monthly y/y (%) |
|---|---|---|
| January to February 2026 | +1.8 | +3.0 |
| March to July 2026 (Destatis), March to October 2026 (IWH) | +8.8 | +11.1 |
| Acceleration (pp) | 7.1 | 8.1 |
| Central acceleration (pp) | 7.6 | |
| Base: March to December 2025, Destatis cases | 20,166 | |
| Implied additional insolvencies, March to December 2026 | about 1,530 |
Source: Destatis; IWH; author's calculations.
Both series show the same pattern. In the first two months of 2026, insolvencies were growing slowly, by about 2 to 3 percent a year. From March onwards, growth accelerated to about 9 to 11 percent. Applying the central acceleration of 7.6 percentage points to the March to December 2025 base gives about 1,530 additional insolvencies in 2026 attributable to the war shock.
5.2 Caveats and cross-checks
The estimate is subject to three caveats. First, official statistics enter with a lag of about three months, so some March and April applications reflect decisions taken before the war; the IWH series, which tracks announcements, is less affected and shows a slightly larger acceleration. Second, the acceleration may contain other factors that intensified after February, including the further rise in Bund yields as inflation expectations increased. Third, the January 2026 increase in the minimum wage of about 8.5 percent could only have affected the January and February figures if its effect had been immediate; the weak growth in those months suggests it was not, but labour cost effects typically build up over a longer period, and some of the post-February acceleration may reflect them.
The industry pattern provides a cross-check. If the war shock works through energy and transport costs, the sectors most exposed to fuel and energy prices should show the strongest acceleration. They do. Insolvency frequency in transport and storage rose from 133 per 10,000 firms in 2025 to an annualised 147 in January to July 2026, an increase of about 10 percent, and in hospitality from 108 to 120, an increase of about 11 percent, against 6 percent for the economy as a whole. In construction, where interest rates matter more than energy, the increase was below average, at about 4 percent. Industry observers quoted in the IWH reporting cite high fuel and staff costs in transport, and energy and staff costs combined with weak consumer sentiment in hospitality. The insolvency administrators’ association VID, by contrast, has argued that the Iran conflict is not an independent driver of insolvencies but an amplifier of pre-existing weaknesses such as outdated business models. Our estimate is consistent with that view: the war shock is real but accounts for less than a fifth of the 2026 excess.
Table 8. Insolvency frequency in the three most affected sectors.
| Sector | 2025, per 10,000 firms | H1 2026 x 2 | Jan to Jul 2026, annualised | Change on 2025 (%) | Relative to economy, Jan to Jul 2026 |
|---|---|---|---|---|---|
| All sectors | 69 | 72.4 | 73.4 | +6.3 | 1.00 |
| Transport and storage | 133 | 143.2 | 146.6 | +10.2 | 2.00 |
| Hospitality | 108 | 119.2 | 120.0 | +11.1 | 1.64 |
| Construction | 104 | 106.8 | 107.8 | +3.7 | 1.47 |
Source: Destatis press releases; author's calculations. Annualisation ignores seasonality and is indicative.
6. Decomposing 2026: How Much Is Cleansing?
6.1 Five components
We can now decompose the estimated 25,800 corporate insolvencies of 2026 into five components, using Model R and the war shock estimate.
The zero-rate baseline is the number of insolvencies Model R predicts if, from 2020 onwards, real GDP had grown at its 2015 to 2019 average of 1.85 percent a year and the Bund yield had stayed at its 2015 to 2019 average of 0.21 percent. It describes a world in which the zero-rate era continued. Rate normalisation is the difference between this baseline and the prediction with actual yields but baseline growth. The demand component is the further difference when actual GDP growth replaces the baseline growth rate. The war and cost shock is the event-study estimate of Section 5. Pandemic catch-up is the remaining excess, up to the remaining deficit in the ledger.
Table 9. Decomposition of 2026 corporate insolvencies.
| Component | Model R (central) | Model T (sensitivity) | Classification |
|---|---|---|---|
| Zero-rate baseline | 17,124 | 13,999 | Baseline |
| Rate normalisation | 5,395 | 1,092 | Cleansing |
| Demand weakness | 962 | 726 | Destruction |
| War and cost shock | 1,531 | 1,531 | Destruction |
| Pandemic catch-up | 766 | 0 | Cleansing |
| Unexplained residual | 0 | 8,430 | Destruction |
| Total 2026 (estimate) | 25,778 | 25,778 | |
| Cleansing (cases) | 6,162 | 1,092 | |
| Destruction (cases) | 2,493 | 10,688 | |
| Cleansing share of the excess (%) | 71 | 9 |
Source: author's calculations.
6.2 Why rate normalisation counts as cleansing
The classification of rate-driven insolvencies as cleansing requires justification, because a firm that fails when its loans are refinanced at higher rates is not necessarily a bad firm. Three arguments support the classification.
First, the level matters more than the change. The yield relevant for 2026 insolvencies, the 2025 average of 2.61 percent, is exactly equal to the 2000 to 2019 average of 2.61 percent and well below the 2000 to 2008 average of 4.25 percent. German firms are not facing exceptional financing costs by historical standards; they are facing normal financing costs after a decade of exceptional ones. A firm that cannot survive at a historically normal cost of capital was being kept alive by the exceptional conditions, not by its own productivity.
Second, this is exactly the mechanism that the literature on zombie firms identifies. When interest rates are very low, firms whose returns no longer cover a normal cost of capital can roll over debt indefinitely, tying up labour and capital that would be more productive elsewhere, and depressing prices and margins for healthy competitors. Their exit when rates normalise is not an accident of the cycle but the resolution of a misallocation.
Third, the composition of the wave fits this interpretation. The rate channel affects firms that borrow, which are predominantly incorporated. Table 3 showed that the 2026 excess is concentrated among partnerships and corporations, while the residual of sole proprietors and very small units is below its pre-pandemic level. The firms failing in larger numbers are therefore the ones the rate channel should affect, rather than the micro-businesses whose failure is driven mainly by local demand and personal circumstances.
The classification is nevertheless a simplification. Some rate-driven insolvencies hit fundamentally sound firms with a temporary liquidity problem, for example in construction and real estate, where the 2022 to 2023 rate shock collided with long project cycles and falling property values. To that extent, our cleansing share is an upper bound. Conversely, our classification of the entire demand component and the entire war shock as destruction is a lower bound for cleansing, since weak demand also eliminates firms that were marginal before the shock. The two biases go in opposite directions.
6.3 Monte Carlo assessment
The decomposition depends on estimated coefficients, on the model choice, on the 2026 outturn and on the war shock estimate. We combine these uncertainties in a Monte Carlo simulation. Each of 20,000 draws selects a model with probability equal to its post-sample likelihood weight (Table 5), draws its coefficients from their estimated joint distribution, draws the 2026 total uniformly between 25,400 and 26,200 and the war acceleration uniformly between 5.6 and 9.6 percentage points. Draws with a persistence coefficient above 0.97 are discarded, leaving 19,980.
Table 10. Monte Carlo results, 19,980 draws.
| Statistic | 10th percentile | Median | 90th percentile | Mean |
|---|---|---|---|---|
| Pandemic deficit 2020 to 2022 (cases) | 5,368 | 8,216 | 11,109 | 8,219 |
| Rate normalisation (cases) | 3,901 | 5,352 | 6,546 | 5,276 |
| Demand weakness (cases) | 260 | 946 | 1,809 | 999 |
| War and cost shock (cases) | 246 | 1,357 | 1,805 | 1,236 |
| Pandemic catch-up in 2026 (cases) | 0 | 0 | 1,542 | 456 |
| Unexplained residual (cases) | 0 | 0 | 3,586 | 842 |
| Cleansing share of the excess (%) | 42 | 70 | 86 | 67 |
| Probability that cleansing exceeds 50 percent | 84% | |||
| Probability of an unexplained residual | 29% |
Source: author's calculations.
The median cleansing share is 70 percent, almost identical to the central decomposition, and the probability that cleansing accounts for more than half of the 2026 excess is 84 percent. The left tail of the distribution, with cleansing shares below 40 percent, comes from draws with a low interest rate coefficient, in which the rate channel is weak and a residual appears that we classify as destruction. The median catch-up in 2026 is zero because, in many draws, the war shock absorbs the whole 2026 excess; the ordering assumption is deliberately conservative.
6.4 Sensitivity
Table 11 shows the cleansing share for combinations of the long-run interest rate semi-elasticity and the war acceleration. The result is robust to the war shock but sensitive to the rate elasticity. With an elasticity of 0.15 or more, the range supported by Models R and C, cleansing accounts for 62 to 80 percent of the excess. With an elasticity of 0.10, the share falls to 57 percent. Only with an elasticity of 0.05, close to the Model T estimate, does destruction dominate.
Table 11. Cleansing share of the 2026 excess (%) under alternative assumptions.
| Long-run rate semi-elasticity | War acceleration 4 pp | 6 pp | 8 pp | 10 pp | 12 pp |
|---|---|---|---|---|---|
| 0.05 | 21 | 21 | 21 | 21 | 21 |
| 0.10 | 57 | 57 | 57 | 57 | 57 |
| 0.15 | 80 | 76 | 71 | 66 | 62 |
| 0.175 (Model R) | 80 | 75 | 70 | 66 | 62 |
| 0.20 | 79 | 74 | 74 | 74 | 74 |
| 0.25 | 88 | 88 | 88 | 88 | 88 |
Source: author's calculations. Constant cells reflect cases in which the war shock is fully absorbed by the model residual or the catch-up stock is exhausted.
7. Does the Wave Reallocate Resources Productively?
7.1 What cleansing requires
A high cleansing share is a necessary but not a sufficient condition for an insolvency wave to raise productivity. Cleansing pays off only if the resources released by failing firms, workers, premises, equipment and market share, move to more productive firms, and if the firms that fail are on average less productive than those that survive. A full assessment would require firm-level productivity data, which are not available in real time. We therefore assess reallocation with five indicators that can be built from published data: the size of failing firms, the composition of the wave, the fate of firms in insolvency, the renewal of the firm stock, and the absorption of displaced workers. Sector productivity statistics and business registration data could not be retrieved for this paper, so the assessment relies on these proxies.
7.2 Who is failing: smaller firms
The average expected creditor claim per corporate insolvency, deflated by the GDP deflator to 2025 prices, is a proxy for the size of failing firms. It fell to about EUR 1.4 million in January to July 2026, compared with EUR 2.7 million in 2024, EUR 2.0 million in 2025 and an average of EUR 1.7 million in 2016 to 2019.
Table 12. Size of failing firms.
| Period | Real claims per case (EUR thousand, 2025 prices) | Employees per case (Destatis) |
|---|---|---|
| 2003 | 1,224 | 5.6 |
| 2009 | 3,295 | 7.7 |
| 2016 to 2019 average | 1,711 | 6.0 |
| 2020 | 3,451 | 11.9 |
| 2021 | 4,166 | 5.4 |
| 2022 | 1,157 | 5.7 |
| 2024 | 2,745 | not published |
| 2025 | 1,991 | not published |
| January to July 2026 | 1,407 | not published |
Source: Destatis; author's calculations. Claims per case are averages and are strongly affected by a few large cases.
The profile is telling. In crisis years dominated by genuine destruction, such as 2009, or by the failure of a few large firms, as in 2020 and 2021, the average claim per case rises sharply. In 2024, large failures in retail, automotive supply and real estate pushed it to EUR 2.7 million. In 2026 it is 18 percent below the pre-pandemic norm. The 2026 wave consists of many small and medium-sized firms, not of large firms toppled by a shock. This is consistent with a cleansing of marginal firms. The 10 percent largest insolvent firms in the IWH sample affected about 54,000 jobs in the first quarter, 45,500 in the second and 41,000 in the third, a declining trend despite rising case numbers.
7.3 What is being destroyed: the industrial core and going-concern value
Two indicators point in the other direction. First, the destruction that does occur falls disproportionately on manufacturing. In October 2026 about 9,400 manufacturing jobs were affected by insolvencies of large firms, the second-highest value since the IWH series began in January 2020. Employment in the car industry is at its lowest level since 2005, with a loss of about 42,300 jobs in a year. These are high-productivity, export-oriented activities whose firm-specific capital is hard to rebuild, and their failures are driven by energy costs, Chinese competition and the transition in automotive technology rather than by zero-rate dependence.
Second, the ability of the insolvency system to preserve going-concern value appears to be deteriorating. According to figures reported in the financial press in July 2026, the success rate of restructurings of larger insolvent firms fell from 57 percent in 2020 to 32.1 percent in 2025. The IWH’s head of insolvency research, Steffen Mueller, summarised the situation in August 2026: the long crisis has apparently used up the reserves of many smaller firms, and market consolidation continues unabated. When the restructuring rate falls, a larger share of every insolvency becomes liquidation, and the value of organisations that could have been saved under new ownership is lost.
7.4 Where the resources go: weak renewal and slow reabsorption
The other side of reallocation is entry and re-employment. Destatis publishes insolvency frequencies per 10,000 firms alongside case numbers, which allows the underlying firm stock to be recovered. It was about 3.49 million in 2025 and about 3.55 million in the January to July 2026 data, an increase of roughly 1.7 percent. The figure depends on how the business register is updated and should be read with caution, but it suggests that the firm population is not shrinking: entry still exceeds exit. On the labour side, the evidence is less favourable. Unemployment has reached about 3.0 million, or 6.3 percent, and the economy is adding public-sector-driven demand rather than private investment. Workers released by failing firms in construction, hospitality and transport can find work in other services, but workers released by industrial firms face a much thinner market for their skills.
Table 13. Reallocation scorecard.
| Indicator | Reading | Signal |
|---|---|---|
| Real claims per insolvency | EUR 1.4 m, 18% below 2016 to 2019 | Cleansing: smaller, marginal firms exit |
| Composition by legal form | Incorporated firms +70 to 80%; residual minus 22% vs 2016 to 2019 | Cleansing: rate-exposed firms exit |
| Pandemic overhang | About 4,200 cases still unreleased | Incomplete cleansing |
| Sector acceleration 2026 | Transport +10%, hospitality +11%, construction +4% | Destruction: cost shock |
| Manufacturing jobs in insolvencies | About 9,400 in October 2026, second-highest since 2020 | Destruction: industrial core |
| Restructuring success of larger firms | 57% (2020) to 32.1% (2025) | Destruction of going-concern value |
| Implied firm stock | About +1.7% from 2025 to 2026 | Weak but positive renewal |
| Unemployment | About 3.0 million (6.3%) | Slow reabsorption |
| Consumer insolvencies | +1.0% in January to July 2026 | No household spiral |
Source: Destatis; IWH; press reports; author's assessment.
The overall verdict is mixed but clear. The selection of firms is broadly healthy: it removes small and medium-sized firms whose business models depended on cheap money, and it does so without a household debt spiral. The reallocation of the resources they release is slow, and the insolvency system is preserving less value than it did five years ago. Germany is getting the cleansing; it is not yet getting the full productivity dividend.
8. Scenarios for 2027
8.1 What is already locked in
Because the insolvency equation uses the previous year’s Bund yield, part of the 2027 outcome is already determined. The average yield in 2026 is about 3.0 percent, 0.4 percentage points above the 2025 average, which adds about 2 percent to 2027 insolvencies on impact and more over time. The minimum wage will rise again, to EUR 14.60 an hour in January 2027, a further increase of about 5 percent. The main open questions are the course of the energy shock and the strength of GDP growth.
We project 2027 with Model R, starting from the 2026 estimate, under three scenarios. In the benign scenario, the Strait of Hormuz reopens, energy prices normalise and GDP grows by 1.5 percent; the war shock partly reverses, which we represent as a 4 percent reduction. In the baseline, GDP grows by 0.9 percent and the 2026 cost level persists. In the adverse scenario, the energy shock intensifies and the economy contracts by 0.5 percent; we add a 6 percent shock, slightly less than the 2026 acceleration. The Monte Carlo bands combine coefficient uncertainty with the equation’s residual variance.
Table 14. Corporate insolvencies in 2027 under three scenarios.
| Scenario | Real GDP growth 2027 (%) | Additional cost shock (%) | Median 2027 | 10th to 90th percentile | Model T (sensitivity) |
|---|---|---|---|---|---|
| Benign | 1.5 | minus 4 | 25,450 | 23,890 to 27,100 | 20,110 |
| Baseline | 0.9 | 0 | 26,670 | 25,050 to 28,390 | 21,120 |
| Adverse | minus 0.5 | +6 | 28,680 | 26,940 to 30,560 | 22,800 |
Source: author's calculations, 20,000 draws per scenario.
8.2 Reading the scenarios
In the baseline, corporate insolvencies rise again, to about 26,700, with a 76 percent probability of exceeding the 2026 level. Even the benign scenario only stabilises the count at about 25,500. The adverse scenario takes insolvencies to about 28,700, with a 69 percent probability of exceeding 28,000, which would be the highest official count since 2011. The Model T sensitivity, in which the pre-2020 downward trend reasserts itself, implies a sharp decline to about 21,000, but this is the model the post-sample evidence rejects.
The scenarios also imply that the composition of the wave will shift. As the catch-up overhang is gradually released and rates stabilise, the cleansing component will plateau; any further increase will come from the cost and demand side. An adverse 2027 would therefore be a year in which destruction, not cleansing, drives the numbers. That is the main reason why the policy response should be designed now, while the wave is still mostly benign.
9. Policy Implications
The analysis supports five conclusions for policy.
First, do not fight the cleansing. About 70 percent of the 2026 excess reflects the exit of firms that could not survive at historically normal financing costs or that were kept alive by pandemic support. A broad rescue programme, a new suspension of filing obligations or generalised loan guarantees would preserve exactly these firms and rebuild the overhang that is only now being worked off. The experience of 2020 to 2022 shows that such measures cannot distinguish viable from unviable firms. They should be reserved for genuinely systemic shocks.
Second, make restructuring faster and cheaper than liquidation. The fall in the restructuring success rate is the most worrying finding of Section 7. Germany’s framework for pre-insolvency restructuring, introduced in 2021, is used far less than its potential. Lower thresholds for access, faster court decisions, specialised insolvency courts with economic expertise and clearer rules for debt-to-equity swaps would allow more viable businesses to survive under new ownership and preserve their organisational capital. For a wave dominated by small and medium-sized incorporated firms, simplified procedures with lower fixed costs are particularly important.
Third, target energy relief narrowly and temporarily. The war shock is the clearest piece of destruction in the 2026 data, and it falls on transport, hospitality and energy-intensive manufacturing. Relief should be limited to firms with demonstrably viable business models, time-limited to the duration of the supply disruption, and designed to preserve the price signal at the margin, for example through lump-sum or capacity-based support rather than price caps. Support that is open-ended or tied to consumption would turn a temporary shock into a new source of zombie firms.
Fourth, invest in reallocation, not in preservation. The payoff of cleansing depends on how quickly workers and capital move to growing firms. Active labour market policy should shift from preserving jobs in failing firms, the logic of short-time work, to supporting transitions: retraining for industrial workers, mobility assistance and faster recognition of qualifications. On the capital side, faster sale of insolvent firms’ assets and simpler business registration would accelerate renewal of the firm stock.
Fifth, monitor the composition, not only the count. The headline number of insolvencies mixes cleansing and destruction. Policymakers should track a small set of compositional indicators, including the share of incorporated firms, real claims per case, manufacturing jobs affected and the restructuring success rate, which together reveal whether a wave is weeding out weak firms or destroying strong ones. Annex C proposes such a dashboard.
10. Conclusion
Germany’s insolvency wave of 2026 has been described as the worst in two decades. Measured by the IWH series of partnerships and corporations, it is. Measured by the official count of all corporate insolvencies, it is a return to the levels of the early 2010s, still a third below the peak of 2003 to 2005. Neither description answers the question that matters: is the wave clearing out firms that should have failed long ago, or destroying firms that should have survived?
Our answer is that it is mostly the former. An insolvency equation estimated only on data from before 2020 predicts the surge of 2023 to 2026 almost exactly, through a single mechanism: the end of zero interest rates. About 70 percent of the 2026 excess over a zero-rate world reflects rate normalisation and the delayed release of pandemic insolvencies, and the probability that cleansing dominates is about 84 percent. The war shock is real, adding about 1,500 cases, but it is an amplifier, not the engine. Half of the pandemic insolvency deficit, about 4,200 firms, has not yet been released.
Four numbers summarise the original findings of this paper: a cleansing share of about 70 percent; a long-run interest semi-elasticity of insolvencies of 0.175, implying about 19 percent more insolvencies per percentage point on the Bund yield; a remaining pandemic overhang of about 4,200 firms; and a baseline projection of about 26,700 corporate insolvencies in 2027. Together they describe a painful but largely healthy adjustment, whose productivity dividend Germany will collect only if it resists the temptation to stop it and invests instead in the speed with which resources move from failing firms to growing ones.
Annex A. Methodology
Data. Annual corporate insolvencies for 1999 to 2022, including employees affected and expected claims, are taken from Destatis, Fachserie 2 Reihe 4.1 (issue 05/2023). 2023 is from Destatis as cited in Bundestag document 20/15080. Monthly corporate and consumer insolvencies and claims for January 2021 to July 2026 are from the Destatis table Insolvenzen nach Monaten (as of 9 October 2026); the 2024 total is the sum of the monthly values (21,812). 2025 annual data and 2026 sector rates are from Destatis press releases. IWH monthly data are taken from IWH press releases and their reproduction in the press. Real GDP growth, ten-year Bund yields, three-month interbank rates, unit labour costs and producer prices of energy are taken from OECD series published by the Federal Reserve Bank of St. Louis (FRED); energy prices for 2023 to 2025 are chained with Destatis annual averages (minus 12.6, minus 6.2 and minus 6.2 percent). The GDP deflator is derived from Destatis nominal GDP growth and real GDP growth.
2026 estimate. January to July official data (15,185) plus August to December 2025 official values multiplied by IWH year-on-year growth for August to October (+9, +13, +9 percent) and an assumed +3 percent for November and December. Range: plus or minus 4 percentage points on monthly growth.
Insolvency equations. OLS on annual data, 2001 to 2019. Model R: log I(t) on a constant, log I(t minus 1), real GDP growth g(t) and the ten-year Bund yield i(t minus 1). Model T adds a linear trend; Model C adds the profit squeeze and real energy price change. Dynamic simulations start from the 2019 actual value and use predicted, not actual, lagged values.
Catch-up ledger. Deficit D is the sum of predicted minus actual insolvencies for 2020 to 2022. From 2023, positive excesses are counted as catch-up until D is exhausted; in 2026 the war shock is attributed first.
Decomposition. Baseline: GDP growth 1.85 percent and Bund yield 0.21 percent from 2020 (2021 for the lagged yield). Rate component: prediction with actual yields and baseline growth minus baseline. Demand component: full prediction minus the rate-only prediction. Cleansing: rate component plus catch-up. Destruction: demand component, war shock and any unexplained residual.
War shock. Difference between year-on-year growth in March to July (Destatis) or March to October (IWH) and January to February 2026, averaged across sources (7.6 percentage points), applied to the March to December 2025 official base of 20,166 cases.
Monte Carlo. 20,000 draws, fixed seed. Model R or T selected with post-sample likelihood weights (effectively 1 and 0). Coefficients from the multivariate normal distribution implied by OLS. 2026 total uniform on 25,384 to 26,172. War acceleration uniform on 5.6 to 9.6 percentage points. Draws with rho above 0.97 discarded (20 draws).
Scenarios. log I(2027) from Model R with actual 2026 estimate as lag, 2026 Bund yield of 3.0 percent, scenario GDP growth and a multiplicative cost shock; bands include residual uncertainty.
Limitations. The estimation sample has 19 annual observations. Annual data cannot capture the within-year timing of filings. Destatis and IWH series differ in coverage and timing. The classification of components into cleansing and destruction is a judgement, discussed in Section 6.2. Business registrations and sector productivity data were not used; reallocation is assessed through proxies. All 2026 figures are estimates until Destatis publishes the full-year data in early 2027.
Annex B. Data Tables
Table B1. Annual data used in the insolvency equations.
| Year | Corporate insolvencies | Real GDP growth (%) | Ten-year Bund yield (%) | Expected claims (EUR m) |
|---|---|---|---|---|
| 1999 | 26,476 | 2.1 | 4.49 | 15,782 |
| 2000 | 28,235 | 2.9 | 5.26 | 21,213 |
| 2001 | 32,278 | 1.6 | 4.80 | 27,223 |
| 2002 | 37,579 | minus 0.2 | 4.78 | 51,759 |
| 2003 | 39,320 | minus 0.5 | 4.07 | 30,522 |
| 2004 | 39,213 | 1.2 | 4.04 | 26,747 |
| 2005 | 36,843 | 0.9 | 3.35 | 22,795 |
| 2006 | 34,137 | 3.9 | 3.76 | 22,239 |
| 2007 | 29,160 | 2.9 | 4.22 | 18,099 |
| 2008 | 29,291 | 0.9 | 3.98 | 22,063 |
| 2009 | 32,687 | minus 5.5 | 3.22 | 73,118 |
| 2010 | 31,998 | 4.1 | 2.74 | 26,594 |
| 2011 | 30,099 | 3.8 | 2.61 | 20,011 |
| 2012 | 28,297 | 0.5 | 1.50 | 41,173 |
| 2013 | 25,995 | 0.5 | 1.57 | 28,061 |
| 2014 | 24,085 | 2.2 | 1.16 | 25,222 |
| 2015 | 23,101 | 1.8 | 0.50 | 17,293 |
| 2016 | 21,518 | 2.3 | 0.09 | 27,376 |
| 2017 | 20,093 | 2.9 | 0.32 | 29,709 |
| 2018 | 19,302 | 1.2 | 0.40 | 21,020 |
| 2019 | 18,749 | 1.1 | minus 0.25 | 26,758 |
| 2020 | 15,841 | minus 4.0 | minus 0.51 | 44,077 |
| 2021 | 13,993 | 4.0 | minus 0.37 | 48,280 |
| 2022 | 14,590 | 1.9 | 1.14 | 14,817 |
| 2023 | 17,814 | minus 1.0 | 2.43 | not available |
| 2024 | 21,812 | 0.0 | 2.32 | about 58,100 |
| 2025 | 24,064 | 0.2 | 2.61 | about 47,900 |
| 2026 (estimate) | about 25,800 | about 1.0 | about 3.0 | 21,900 (January to July) |
Source: Destatis; OECD via FRED; 2026 GDP: author's estimate from quarterly data (Q3 and Q4 at plus 0.2 percent).
Table B2. Annual sums of the monthly insolvency table.
| Year | Corporate (sum of months) | Consumer | All claims, monthly table (EUR m) |
|---|---|---|---|
| 2021 | 13,993 | 79,620 | 55,910 |
| 2022 | 14,590 | 66,428 | 31,005 |
| 2023 | 17,814 | 66,887 | 33,633 |
| 2024 | 21,812 | 71,207 | 66,787 |
| 2025 | 24,064 | 77,219 | 56,535 |
| January to July 2026 | 15,185 | 46,043 | 27,538 |
Source: Destatis, Insolvenzen nach Monaten. Claims in this table cover all insolvency proceedings, not only corporate ones.
Annex C. An Insolvency Quality Dashboard
We propose six indicators, all published at least quarterly, to monitor whether the wave remains predominantly cleansing.
Table C1. Insolvency quality dashboard. Thresholds: author's proposal.
| Indicator | Source and frequency | Warning threshold (destruction) | Current reading | Status |
|---|---|---|---|---|
| IWH excess over 2016 to 2019 versus Destatis excess | Monthly, IWH and Destatis | Gap closing while both rise | 79% versus 29% | Cleansing |
| Real claims per case | Monthly, Destatis | Above EUR 2.5 m | EUR 1.4 m | Cleansing |
| Manufacturing jobs in large insolvencies | Monthly, IWH | Above 8,000 for three months | 9,400 (October) | Warning |
| Insolvency rate, transport and hospitality relative to all sectors | Monthly, Destatis | Rising relative rate | Rising | Warning |
| Restructuring success of larger firms | Annual, industry studies | Below 40% | 32.1% (2025) | Warning |
| Unemployment | Monthly, Federal Employment Agency | Above 3.0 million sa | About 3.0 million | Watch |
Three of the six indicators were flashing a warning in the autumn of 2026. The selection of firms remains healthy; the preservation of value and the reallocation of resources are where Germany is losing ground.
References
Boersen-Zeitung (2026). Firmenpleiten auf 21-Jahreshoch. July 2026.
Boerse Express (2026). Unternehmensinsolvenzen: Deutschland erreicht Rekord seit 21 Jahren. 26 July 2026.
Deutscher Bundestag (2020). Drucksache 19/18364, Antwort der Bundesregierung (insolvency statistics 2010 to 2019).
Deutscher Bundestag (2025). Drucksache 20/15080 (insolvency statistics 2023 and 2024).
Destatis (2020). Number of business insolvencies down 2.9% in 2019 year on year. Press release 094/2020.
Destatis (2023). Fachserie 2 Reihe 4.1, Insolvenzverfahren, May 2023 (annual series 1999 to 2022).
Destatis (2024, 2025, 2026). Producer price press releases: December 2023 (030/2024), December 2024 (January 2025), December 2025, August 2026 (328/2026), September 2026 (373/2026).
Destatis (2026). Insolvenzen nach Monaten, table as of 9 October 2026.
Destatis (2026). Unternehmensinsolvenzen im Juni 2026: +15,8 % gegenueber Juni 2025. Press release 325/2026; press releases on 2025 annual results and on July 2026 data.
Federal Reserve Bank of St. Louis (2026). FRED series IRLTLT01DEA156N, IR3TIB01DEA156N, NAEXKP01DEA657S, CLVMNACSCAB1GQDE, ULQEUL01DEQ659S, PIEAEN02DEA661N (OECD and Eurostat data).
Destatis (2026). Volkswirtschaftliche Gesamtrechnungen, Verwendung des Bruttoinlandsprodukts in jeweiligen Preisen, long series.
IMK (2026). Policy Brief 208, Inflation monitor, January 2026.
Leibniz-Institut fuer Wirtschaftsforschung Halle (2026). IWH-Insolvenztrend, monthly releases January to October 2026, including 9 July 2026 (second quarter), 6 August 2026 (July) and 5 November 2026 (October).
Mindestlohnkommission (2025). Decision on the statutory minimum wage for 2026 and 2027.
Statistiken-aktuell (2026). Unternehmensinsolvenzen September 2026, Datenmonitor.
TRANSPORT (2026). Insolvenzen: Keine Branche trifft es haeufiger als Verkehr und Lagerei. October 2026.
Driss EFFINA, PhD
Independent Researcher, Global Strategy Files
The opening pages of our four country reports: United States, United Kingdom, Japan and China. A free PDF, sent to your inbox.
You will also receive occasional analysis from Global Strategy Files. Unsubscribe in one click, at any time.