Decoded, Economics & finance

Survivor Elasticity: Why Germany’s Second Energy Shock Hits a Different Industry Than 2022

Survivor Elasticity
The German Economy in 2026 | Analytical Series, Paper 2

Executive Summary

Germany is living through its second energy shock in four years. The US and Israeli war on Iran, which began on 28 February 2026, disrupted the Strait of Hormuz and forced Qatar to declare force majeure on liquefied natural gas deliveries. The front-month TTF gas price, which stood under 32 EUR/MWh at the end of February, averaged 51 EUR/MWh between April and September and reached 85 EUR/MWh for November delivery. German storage was 61.4 percent full on 1 November, a record low for the start of a heating season and well short of the effective target of about 70 percent. The obvious question is whether German energy-intensive industry is heading for a repeat of 2022, when its output fell by about a sixth within a year.

This paper argues that the answer depends on recognising that the 2026 shock is hitting a different industry. The 2022 shock did not simply depress production for a while. It removed the most gas-sensitive capacity from the German industrial base. What remains is a survivor population: plants that withstood gas prices of more than 200 EUR/MWh, or that were restructured, hedged or repurposed to do so. A survivor population should respond differently to a new price shock. We call this property survivor elasticity, and we measure it.

Our first finding is that survivor elasticity is low in the short run. Using monthly seasonally adjusted production data from the Bundesbank for September 2021 to September 2026, we compare the response of chemicals, metals and intermediate goods to the two price shocks, controlling for total industrial output. In the second half of 2022, a 48 log-point rise in the TTF price was associated with a fall of 9.3 percent in chemical output, an elasticity of about minus 0.19. In the third quarter of 2026, a 55 log-point rise was associated with a fall of 2.6 percent, an elasticity of about minus 0.05. For metals and intermediate goods the 2026 response is statistically indistinguishable from zero.

Our second finding explains why. The 2022 loss behaved like a ratchet. A hysteresis model in which relative output depends on the highest six-month average price reached so far, plus a small contribution from the current price, fits the 2022 to early-2026 data well and explains almost the entire 2022 loss in relative chemical output through the ratchet term. Estimated only on data up to February 2026, it forecasts the 2026 path out of sample with a root mean squared error below 2 points. The decomposition is stark: the energy-intensive aggregate fell by 17 percent between February and December 2022, recovered only 8 percent of that loss by 2025 when gas prices had returned to 34 to 38 EUR/MWh, and was back at its trough by the third quarter of 2026.

The ratchet has a reassuring and a worrying side. The reassuring side is that the surviving capacity will be lost permanently only if the six-month average gas price climbs above the level at which the ratchet sits, which we estimate at 136 to 164 EUR/MWh, more than twice the 61 EUR/MWh average for June to November 2026. The worrying side is that the threshold is not fixed. For chemicals it is eroding by about 5 percent a year, and the energy-intensive branches now employ about 75,000 more people than their output would justify at 2022 productivity. The survivors are less elastic, but they are also thinner and more fragile at the margin.

The binding risk for this winter is therefore not the price level as such but the physical gas balance. We build a Monte Carlo model of German storage for November 2026 to March 2027, calibrated on the storage association’s normal-winter scenario, the observed withdrawal of winter 2025/26 and its very cold scenario, with an explicit supply stress component. Starting from 61.4 percent, the probability that storage is below the 30 percent legal floor on 1 February 2027 is 38 percent, and the probability that it is below 20 percent is 18 percent. A replay of last winter’s weather would leave storage at 22 percent on 1 February and about 6 percent in early March.

Translating storage outcomes into prices, curtailment and permanent closures, we find that the median loss of energy-intensive output over the winter is modest, at 1.1 percent of annual output. But the distribution has a long tail: the mean loss is 4.4 percent, and in the worst 5 percent of outcomes the loss exceeds 22 percent, comparable to 2022. The expected GDP cost is 0.18 percentage points, with a 5 percent tail of 0.9 points. Had the industry retained its 2022 elasticity, the expected cost would be almost three times higher. The policy conclusion follows directly: Germany’s priority for the winter should be to protect the gas balance and the survivors’ threshold, through demand-side flexibility, procurement of additional LNG and early, transparent curtailment rules, rather than broad price subsidies for an industry whose short-run output is no longer very sensitive to price.

Key findings.

  1. The industry that faces the 2026 shock responds four to five times less to gas prices than the industry of 2022. The arc elasticity of chemical output to the TTF gas price was about minus 0.19 in the second half of 2022, against about minus 0.05 in the third quarter of 2026. For metals and intermediate goods, the 2022 response of minus 0.09 to minus 0.14 has turned into a small positive number in 2026.
  2. The 2022 loss was a ratchet, not a temporary curtailment. A hysteresis model in which relative output depends on the highest six-month average gas price reached so far explains 98 percent of the 2022 loss in relative chemical output and forecasts 2026 out of sample with an error of under 2 points. The energy-intensive aggregate recovered only 8 percent of its 2022 loss by 2025 and was back at its 2022 trough, 17 percent below February 2022, in the third quarter of 2026.
  3. Survivors break only above 136 to 164 EUR/MWh, but that threshold is eroding. Permanent losses resume when the six-month average gas price exceeds the old peak, against 61 EUR/MWh for June to November 2026. For chemicals the threshold is falling by about 5 percent a year, and the energy-intensive branches carry a labour overhang of roughly 75,000 jobs relative to 2022 productivity.
  4. Storage is the binding risk, not prices. Starting from 61.4 percent on 1 November 2026, our Monte Carlo model gives a 38 percent probability that storage is below the 30 percent legal floor on 1 February 2027 and an 18 percent probability that it is below 20 percent. A simple replay of last winter's weather would leave storage at 22 percent on 1 February and about 6 percent in early March.
  5. Output at risk has a narrow centre and a long tail. The median loss of energy-intensive output over the winter is 1.1 percent of annual output, but the mean is 4.4 percent and the worst 5 percent of outcomes exceed 22 percent, a repeat of 2022. In GDP terms the expected cost is 0.18 percentage points, with a 5 percent tail of 0.9 points.
  6. The survivor dividend is worth about one third of a point of GDP, and fuller storage would halve the expected loss. Had the industry kept its 2022 elasticity, the expected output loss would be 12.2 percent rather than 4.4 percent. Starting the winter at the 77 percent that the storage association considered technically feasible would have cut the expected loss to 2.1 percent.

1. Introduction: The Second Shock Meets a Different Industry

When Russian gas deliveries collapsed in 2022, German energy-intensive industry became the shock absorber of the national gas balance. Production in the five energy-intensive branches (chemicals, metals, glass and ceramics, paper, and coke and refined petroleum) fell by more than 17 percent between the start and the end of 2022, according to an analysis by the ifo Institute. Chemical production fell by 27 percent between December 2021 and December 2022 on ifo’s measure. Gas demand fell, storage was filled, households were protected, and the feared rationing never happened. In a narrow sense, the policy of letting high prices clear the market worked.

The cost of that success became visible only slowly. By March 2026, Destatis reported that production in the energy-intensive branches was still 15.2 percent below its February 2022 level, against 9.5 percent for industry as a whole. Glass, ceramics and stone products were down 25.0 percent, paper 18.5 percent, chemicals 18.1 percent and metals 12.9 percent. Only mineral oil processing had grown, by 24.6 percent, mostly since January 2026. Employment in the energy-intensive branches had fallen from 847,700 to 794,400. The industry that emerged from the 2022 shock was smaller, and it did not grow back when gas prices fell.

Then came the second shock. The war on Iran, which began on 28 February 2026, closed or disrupted the Strait of Hormuz and removed a large share of Qatar’s LNG exports from the world market. Qatar accounts for roughly 20 percent of global LNG export capacity. On 18 March 2026, 17 percent of the capacity of Ras Laffan was knocked out for a period of up to five years, according to reports compiled by the energy publication Cleanthinking. The TTF gas price rose by more than 85 percent in the first trading days of March, and German storage, which had fallen to 21 percent in early March after a cold January, reached its lowest September level in 15 years of records.

The natural reflex is to read 2026 through the lens of 2022: higher gas prices, falling energy-intensive output, more plant closures, and a further step down in Germany’s industrial base. This paper argues that the reflex is partly wrong, and that being precise about where it is wrong matters for policy. The 2026 shock hits an industry that has already shed its most gas-sensitive capacity. The marginal plant of 2022, for example an unhedged unit with thin margins in basic chemicals, is gone. The marginal plant of 2026 is one that survived prices of more than 200 EUR/MWh. A survivor population should, in the short run, respond much less to a given price increase. But it may also be closer to an irreversible threshold, because the shock it survived weakened its balance sheet, its order book and its competitive position.

We call the short-run response of this survivor population survivor elasticity, and the paper is an attempt to measure it and to draw the consequences for the winter of 2026/27. Our central hypothesis has two parts. First, the output elasticity of energy-intensive production to gas prices is lower in 2026 than in 2022, because the 2022 shock worked through a ratchet that permanently removed the most elastic capacity. Second, the main risk in 2026 is not a gradual price-driven decline but a discontinuous event: a physical shortfall of gas late in the winter, or a price spike large enough to push the six-month average above the survivors’ threshold, which would trigger a new round of permanent closures.

The paper proceeds as follows. Section 2 compares the two shocks in terms of prices, storage and the state of the industrial base. Section 3 estimates arc elasticities for the two episodes. Section 4 introduces a hysteresis model that separates permanent ratchet effects from temporary price effects and tests it out of sample on 2026. Section 5 decomposes the 2022 to 2026 loss of output into its reversed and permanent parts and introduces a labour overhang indicator. Section 6 presents a Monte Carlo model of German gas storage for winter 2026/27. Section 7 translates storage and price outcomes into energy-intensive output and GDP at risk. Section 8 draws policy implications, and Section 9 concludes. An annex sets out the data, the methods and their limitations.

2. Two Shocks Compared

2.1 The price shock: smaller in level, similar in proportion

The two shocks look very different in levels. Table 1 summarises the TTF month-ahead index, published monthly by the European Energy Exchange and applied to each delivery month. In 2022 the index averaged 132 EUR/MWh, with a peak of 234 EUR/MWh for September delivery. In 2026 it rose from about 31 EUR/MWh in January and February to an average of 51 EUR/MWh between April and September, and then to 75 EUR/MWh for October and 85 EUR/MWh for November delivery.

Table 1. TTF gas prices in the two shock episodes.

Period TTF month-ahead index, average (EUR/MWh) Note
Jan to Feb 2022 100.2 Base of the 2022 shock window
Jul to Dec 2022 162.0 2022 shock window
Aug to Oct 2022 (3-month peak) 203.0 Highest three-month average
2022, full year 131.8
2023 48.5
2024 33.9
2025 38.0
Jan to Feb 2026 30.6 Base of the 2026 shock window
Apr to Sep 2026 50.7 2026 shock window
Jul to Sep 2026 53.3 Third quarter 2026
Jun to Nov 2026 (6-month average) 61.3 Latest six-month average
November 2026 delivery 85.3 Latest value

Source: EEX monthly TTF index as published by Protergia; author's calculations.

050100150200250Jan 22Jan 23Jan 24Jan 25Jan 26EUR/MWhSep 2022: 234Nov 2026: 852022 shock2026 shock
Figure 1. TTF month-ahead gas price index by delivery month, January 2022 to November 2026. Source: EEX via Protergia; author's chart.

In proportional terms the two shocks are closer than the levels suggest. Between the base period and the shock window, the log price rose by 0.48 in 2022 and by 0.50 in 2026; for the third quarter of 2026 alone the rise was 0.55. This is convenient for our purposes. It means that the two episodes can be compared on an equal footing as proportional shocks, which is how firms experience them: a doubling of the gas bill matters in roughly the same way whether it starts from 30 or from 100 EUR/MWh, at least for firms whose margins were calibrated on the pre-shock price.

The two shocks differ in two other respects. First, in 2022 the base itself was already elevated, at about 100 EUR/MWh, roughly three times the 2024 to 2025 average, whereas the 2026 shock started from the lowest prices since the spring of 2024. Second, the 2026 shock is accompanied by a much weaker storage position, which we discuss in Section 6. In 2026, according to the storage association INES as reported by the press, injecting gas into storage was loss-making for much of the summer, and filling was correspondingly slow.

2.2 The industrial base after 2022

The second difference is the industry itself. Figure 2 shows the production indices published by the Bundesbank for the two largest energy-intensive groups that it reports separately, chemicals and pharmaceuticals (WZ 20 and 21) and metals and metal products (WZ 24 and 25), together with intermediate goods and total industry. All series are seasonally adjusted, with 2021 equal to 100.

80859095100105Jan 22Jan 23Jan 24Jan 25Jan 26Index, 2021 = 100Industry, totalMetalsIntermediate goodsChemicals and pharma
Figure 2. Production in selected industries, September 2021 to September 2026 (2021 = 100, seasonally adjusted). Source: Deutsche Bundesbank, Saisonbereinigte Wirtschaftszahlen, November 2026.

Three features stand out. Chemicals fell first and fastest in 2022, from about 100 in early 2022 to below 88 by December. Metals and intermediate goods fell more slowly, but continued to decline through 2023 and 2024, long after gas prices had normalised. Total industry, by contrast, held up in 2022 and only began its own decline in 2023, under the combined effect of weak external demand, the automotive crisis and higher interest rates. By September 2026, total industry stood at 92.4, chemicals and pharmaceuticals at 84.7, metals at 86.4 and intermediate goods at 83.8.

The branch structure of the loss confirms that gas intensity, not general weakness, is the dominant explanation. Table 2 reproduces the Destatis comparison between February 2022 and March 2026, which includes the smaller branches not reported separately by the Bundesbank.

Table 2. Production and employment in the energy-intensive branches.

Branch Production, Feb 2022 to Mar 2026 (%) Employment change, Feb 2022 to Mar 2026
Concrete, cement and lime-sandstone products minus 29.3 not reported
Glass, ceramics, stone and earth products minus 25.0 minus 6.4% (minus 9,800)
Paper industry minus 18.5 minus 8.6% (minus 10,200)
Chemical industry minus 18.1 minus 5.5% (minus 18,300)
Metal production and processing minus 12.9 minus 7.1% (minus 16,000)
Mineral oil processing plus 24.6 plus 5.8% (plus 1,000)
Energy-intensive branches, total minus 15.2 minus 6.3% (minus 53,200)
Industry, total minus 9.5 not reported

Source: Destatis, as reported by Produktion and onvista, May 2026.

The table also reveals something that will matter in Section 5: in every branch except mineral oil processing, employment fell far less than output. The energy-intensive branches lost 15.2 percent of their output but only 6.3 percent of their workforce. That gap is a measure of labour hoarding, and it is one of the reasons we think the survivors are more fragile than their low short-run elasticity suggests.

2.3 What the 2026 production data show so far

Destatis publishes the production index of the energy-intensive branches every month. Table 3 collects the first releases for 2026. The pattern is not one of collapse. Production rose by 1.9 percent in February, 1.2 percent in March and 1.0 percent in April, then fell in June, July and August, and rose again by 0.8 percent in September. In year-on-year terms the index swung between minus 4.3 percent in January and plus 2.9 percent in May, before settling at about minus 2 percent in August and September. In the third quarter, production was 3.5 percent lower than in the second.

Table 3. Monthly production in the energy-intensive branches in 2026, first releases.

Month (2026) Energy-intensive branches, m/m (%) Energy-intensive branches, y/y (%) Total production, m/m (%)
January minus 0.8 minus 4.3 minus 0.5
February plus 1.9 plus 0.1 minus 0.3
March plus 1.2 minus 1.2 minus 0.7
April plus 1.0 plus 0.9 plus 0.4
May plus 0.2 plus 2.9 plus 0.9
June minus 1.8 plus 2.2 plus 0.2
July minus 1.7 minus 0.5 minus 1.1
August minus 0.5 minus 2.1 plus 2.0
September plus 0.8 minus 1.9 plus 0.7

Source: Destatis press releases; total production refers to industry including energy and construction.

Part of the spring resilience reflects a composition effect. Mineral oil processing, which is counted among the energy-intensive branches, expanded strongly from January 2026 as refinery margins rose with the war. Chemicals, the most gas-intensive large branch, show a clearer reaction: the Bundesbank index for chemicals and pharmaceuticals fell from 85.8 in June to 81.0 in July and 80.8 in August, its lowest level in the series, before recovering to 84.7 in September. The third-quarter average of 82.2 was 4.0 percent below the second quarter. The question is how large this reaction is relative to the size of the price shock, and how it compares with 2022.

3. Measuring the Response: Episode Elasticities

3.1 Method

The simplest way to compare the two shocks is an arc elasticity: the change in log output between a base period and a shock window, divided by the change in the log gas price over the same periods. For 2022 we take January and February 2022 as the base and July to December 2022 as the shock window. For 2026 we take January and February 2026 as the base and both April to September 2026 and the third quarter of 2026 as alternative windows. The windows are chosen to start one to two months after the price increase, because industrial gas is largely bought forward and the effective cost of gas reaches firms with a lag.

We compute the elasticity in two forms. The absolute elasticity uses the raw change in branch output. The relative elasticity uses the change in branch output relative to total industry, which removes common demand shocks such as the global industrial cycle or the automotive crisis. Because both forms rely on a handful of monthly observations, we attach a standard error derived from the month-to-month noise of relative output in the calm years 2024 and 2025.

3.2 Results

Table 4. Arc elasticities of output to the TTF gas price in the two shocks. Standard errors from calm-period noise: about 0.03 for chemicals and metals, 0.014 for intermediate goods.

Branch and episode Change in log price Output (%) Relative output (%) Elasticity, absolute Elasticity, relative
Chemicals, 2022 (Jul to Dec) 0.48 minus 9.3 minus 8.2 minus 0.19 minus 0.17
Chemicals, 2026 (Apr to Sep) 0.50 minus 0.5 minus 0.2 minus 0.01 0.00
Chemicals, 2026 (Q3) 0.55 minus 2.6 minus 1.9 minus 0.05 minus 0.03
Metals, 2022 (Jul to Dec) 0.48 minus 4.3 minus 3.2 minus 0.09 minus 0.07
Metals, 2026 (Q3) 0.55 plus 0.4 plus 1.1 plus 0.01 plus 0.02
Intermediate goods, 2022 (Jul to Dec) 0.48 minus 6.7 minus 5.6 minus 0.14 minus 0.12
Intermediate goods, 2026 (Q3) 0.55 plus 0.4 plus 1.1 plus 0.01 plus 0.02

Source: Bundesbank, EEX via Protergia; author's calculations.

The contrast is sharp. In 2022, chemical output fell by about 0.19 percent for every 1 percent increase in the gas price, and by 0.17 percent relative to industry. In 2026, even in the weakest quarter so far, the elasticity is a quarter of that in absolute terms and a fifth in relative terms. For metals and intermediate goods the 2026 elasticity is slightly positive, and well within one standard error of zero. The 2022 elasticities for chemicals and intermediate goods are more than four standard errors from zero; the 2026 elasticities are not significantly different from zero for any of the three groups.

889296100104M3M4M5M6M7M8M9M10M11Relative output, base = 1002022 shock (base Jan to Feb 2022)2026 shock (base Jan to Feb 2026)
Figure 3. Relative chemical output in the two shock years, indexed to January and February of each year (base = 100). Source: Bundesbank; author's calculations.

Figure 3 shows the same contrast month by month. In 2022, relative chemical output fell steadily from April onwards and ended the year about 11 percent below its base. In 2026 it held above the base until June, dipped by about 3 percent in July and August, and returned to the base in September. The 2026 path looks like a temporary adjustment of utilisation; the 2022 path looks like a change of level.

3.3 Caveats

Three caveats apply. First, the 2026 window is short. The shock has lasted seven months, and the elasticity of 2022 might also have looked small after seven months had prices not risen further in the summer. We address this in Section 4 by modelling the dynamics explicitly. Second, the Bundesbank series combines chemicals with pharmaceuticals, which are not energy-intensive and which have their own cycle; pharmaceutical output rose by 8.0 percent in September 2026 alone. This dilutes the measured response of chemicals proper in both episodes, but it does not explain why the response differs between them. Third, the month-ahead index is a proxy for the effective gas cost of firms. Many large consumers buy gas one to three years forward, and the 2022 effective cost lagged the spot market by months. If anything, this biases the 2026 comparison towards finding a larger response, because firms that hedged at the low prices of 2024 and 2025 have been partly protected so far.

None of these caveats is large enough to close the gap. To bring the 2026 chemical elasticity up to its 2022 level, chemical output would have had to fall by a further 8 percent in the third quarter, to around 76 on the 2021 base, a level never reached in the series.

4. A Ratchet, Not a Curve

4.1 The model

Why should the same industry respond so differently to proportionally similar shocks? The economic intuition comes from the theory of investment under uncertainty with sunk costs. Plants differ in the gas price at which they cease to cover their variable costs. When the price rises above a plant’s threshold for long enough, the plant is mothballed, and because restarting requires new investment, new staff and new customers, it often never returns. Once the plants with the lowest thresholds have exited, the next shock of the same size finds no one to remove. Output then depends not on the current price, but on the highest sustained price the industry has experienced. This is the logic of a ratchet.

We formalise it with a simple hysteresis regression. Let r(t) be the log of branch output relative to total industry, multiplied by 100. Let P(t) be the six-month moving average of the TTF price, a proxy for the gas cost that firms actually pay, and let M(t) be the running maximum of P since January 2022. We estimate:

r(t) = a + b x log M(t) + c x log P(t) + error

The coefficient b measures the permanent ratchet: how much relative output is lost, for good, when the maximum sustained price rises. The coefficient c measures the temporary response to the current price, which disappears when prices fall. If the 2022 loss was a ratchet, b should be large and c small. We estimate the model on January 2022 to February 2026, before the Iran war, and then use it to forecast the period March to September 2026 out of sample.

We also estimate a variant in which the ratchet erodes over time: the running maximum is allowed to decay at a rate d per year, so that a plant that survived a high price in 2022 may become vulnerable to a lower price later, as its margins are eroded by competition, carbon costs or ageing equipment. The decay rate is chosen to minimise the sum of squared residuals within the estimation sample.

4.2 Results

Table 5. Hysteresis model of relative output, six-month price average, no erosion. Estimation sample January 2022 to February 2026 (50 months); HAC standard errors in brackets.

Branch Ratchet b Current price c R squared Ratchet share of 2022 loss (%) RMSE, Mar to Sep 2026
Chemicals and pharma minus 29.0 (6.7) minus 3.5 (0.9) 0.50 98 1.76
Metals minus 14.6 (2.0) plus 0.5 (0.2) 0.61 over 100 0.83
Intermediate goods minus 22.3 (2.1) plus 0.4 (0.2) 0.86 70 1.38

Source: author's estimates on Bundesbank and EEX data.

The ratchet coefficient is large, negative and highly significant in all three branches. For chemicals, each 1 percent increase in the maximum sustained gas price permanently lowered relative output by about 0.29 percent; for intermediate goods by 0.22 percent and for metals by 0.15 percent. The current-price coefficient, by contrast, is small: minus 0.035 for chemicals and slightly positive for metals and intermediate goods. In other words, once the ratchet is accounted for, the temporary response of output to gas prices is close to zero outside chemicals, and modest within chemicals.

The decomposition of the 2022 loss is the most striking result. Between January and December 2022, relative chemical output fell by 9.9 log points. The ratchet term accounts for 9.7 of them, or 98 percent; the current-price term for about 1.2 points, with the residual slightly offsetting. For metals the ratchet term more than accounts for the observed relative loss. For intermediate goods it accounts for 70 percent, the remainder reflecting the sharp drop of December 2022. The 2022 shock was, overwhelmingly, a permanent event.

-20-15-10-505Jan 22Jan 23Jan 24Jan 25Jan 26log ratio x 100out of sampleActualFitted: ratchet plus current priceRatchet component
Figure 4. Hysteresis model of relative chemical output: actual values, fitted values and the ratchet component, January 2022 to September 2026. The model is estimated up to February 2026. Source: author's estimates.

4.3 The out-of-sample test: 2026

The most demanding test of the model is the 2026 shock, which it has not seen. Because the six-month average price in 2026 has remained far below its 2022 maximum, the model predicts that only the small current-price term should be active. For chemicals it predicts a decline in relative output of about 1.4 points between January and February 2026 and the third quarter; the actual decline was 1.9 points. For metals and intermediate goods it predicts almost no change; the actual changes were small and positive. The out-of-sample root mean squared errors, between 0.8 and 1.8 points of relative output, are of the same order as the month-to-month noise of the series.

The survivor elasticity of Section 3 is therefore not an accident of timing. It is what a ratchet model, estimated on the pre-war period, predicts for a shock that does not exceed the previous maximum. The 2022 elasticity was high because the 2022 shock pushed the price above the thresholds of a large part of the capacity. The 2026 elasticity is low because, so far, it has not.

4.4 Where the ratchet sits now

The policy-relevant question is how far the current price is from the point at which the ratchet would bite again. Table 6 reports the estimated threshold, that is the level of the six-month average price above which the model predicts new permanent losses, under alternative specifications.

Table 6. Robustness of the hysteresis model and the implied ratchet threshold.

Specification Erosion (per year) Ratchet b Current price c Threshold, Sep 2026 (EUR/MWh)
Chemicals, 6-month average, no erosion 0% minus 29.0 minus 3.5 164
Chemicals, 6-month average, best-fit erosion 5% minus 29.9 minus 1.6 136
Chemicals, 3-month average, no erosion 0% minus 16.1 minus 2.0 203
Chemicals, 3-month average, best-fit erosion 11% minus 17.5 plus 0.4 129
Metals, 6-month average, best fit 0% minus 14.6 plus 0.5 164
Intermediate goods, 6-month average, best fit 0% minus 22.3 plus 0.4 164

Source: author's estimates.

For metals and intermediate goods, the data prefer a ratchet without erosion, located at the 2022 peak of the six-month average price, 164 EUR/MWh. For chemicals, the data prefer an eroding ratchet, at about 5 percent a year with six-month averaging and 11 percent with three-month averaging, which places the threshold at 129 to 136 EUR/MWh in September 2026. The six-month average for June to November 2026 is 61 EUR/MWh. The survivors are therefore not close to their threshold under current prices. To trigger a new round of permanent losses, the six-month average would need to more than double, which would require monthly prices well above 150 EUR/MWh for most of the winter.

The erosion estimate for chemicals deserves emphasis, because it is the quantitative core of the “fragile survivor” argument. A threshold that falls by 5 percent a year halves in about 14 years; in four years it moves from 164 to about 134 EUR/MWh. Chemicals have faced a structural squeeze since 2022 that has nothing to do with the gas price directly: overcapacity in Chinese basic chemicals, the carbon price, weak domestic demand from construction and automotive customers. Each of these lowers the gas price at which a plant stops being viable. The ratchet that protects the survivors is real, but it is sinking.

5. The Survivor Decomposition: What 2022 Removed

5.1 The aggregate

If the 2022 loss was mainly temporary, output should have recovered when gas prices fell to the 30 to 40 EUR/MWh range in 2024 and 2025. We test this directly with a survivor decomposition. For each series we measure three things: the shock loss, from the pre-war base to the trough at the end of 2022; the loss remaining in 2025, when gas prices were close to their pre-2021 levels; and the recovery ratio, the share of the shock loss that had been reversed by 2025.

For the energy-intensive aggregate, the Destatis index is not available to us as a continuous series for 2022 to 2024, so we reconstruct its key points from official releases. The index stood at about 98.8 in February 2022, implied by the 15.2 percent decline to the March 2026 level of 83.8. The September 2022 release reported a year-on-year decline of 9.7 percent, followed by monthly declines of 3.6 percent in October and 6.1 percent in December and a small rise in November. Chained from September, this places December 2022 at about 81.9, consistent with ifo’s estimate of a fall of more than 17 percent during 2022 and with the 19.6 percent year-on-year decline reported for December. The annual changes of plus 0.6 percent in 2024 and minus 2.6 percent in 2025, together with the 2026 monthly releases, complete the series.

Table 7. Survivor decomposition of energy-intensive output, 2022 to 2026 (2021 = 100). Pre-war base: average of January and February 2022 (February 2022 for the aggregate). Trough: fourth quarter of 2022 (December 2022 for the aggregate). 2025 for the aggregate: January to September average.

Series Pre-war Trough 2025 Q3 2026 Shock loss (%) Loss in 2025 (%) Recovery (%)
Energy-intensive (5) 98.8 81.9 83.3 81.8 minus 17.1 minus 15.7 8
Chemicals and pharma 100.0 89.5 86.3 82.2 minus 10.5 minus 13.7 minus 30
Metals 100.3 95.4 86.8 85.8 minus 4.8 minus 13.4 minus 179
Intermediate goods 100.4 92.2 84.7 83.8 minus 8.2 minus 15.6 minus 90
Industry, total 100.8 99.8 93.2 91.1 minus 0.9 minus 7.5 not meaningful

Source: Destatis, Bundesbank; author's calculations.

The answer is unambiguous. The energy-intensive aggregate recovered only 8 percent of its 2022 loss by 2025, despite gas prices having fallen by more than two thirds from their 2022 average. For chemicals, metals and intermediate goods the recovery ratio is negative: their output in 2025 was lower than at the trough of 2022. By the third quarter of 2026, the energy-intensive aggregate was back at its December 2022 level, 17.2 percent below February 2022.

-20-15-10-50Energy-intensiveChem. and pharmaMetalsIntermediateIndustry% vs Jan to Feb 2022Shock: to Q4 2022To 2025 averageTo Q3 2026
Figure 5. Survivor decomposition: percentage change from the pre-war base to the end-2022 trough, to the 2025 average and to the third quarter of 2026. Source: Destatis, Bundesbank; author's calculations.

5.2 Separating the gas shock from the industrial cycle

Part of the continued decline after 2022 reflects the general weakness of German industry rather than the gas shock. Total industry fell by 7.5 percent between the pre-war base and 2025, mostly in 2023 and 2024. The relative measures in Table 8 strip out this common component.

Table 8. Survivor decomposition relative to total industry.

Branch At end-2022 trough (%) In 2025 (%) In Q3 2026 (%) Recovery by 2025 (%)
Chemicals and pharma minus 9.7 minus 6.7 minus 9.2 31
Metals minus 3.9 minus 6.4 minus 5.4 minus 63
Intermediate goods minus 7.4 minus 8.8 minus 7.8 minus 20

Source: Bundesbank; author's calculations.

Relative to industry, chemicals did recover part of their loss, about 31 percent by 2025, largely because the rest of industry caught up with them on the way down. By the third quarter of 2026 that partial relative recovery had been almost entirely lost again. Metals and intermediate goods show no relative recovery at all. Whichever way the numbers are cut, between 70 and 100 percent of the 2022 gas shock loss must be classified as permanent.

5.3 The labour overhang: a fragility indicator

The survivor population is less elastic, but is it healthy? One indicator suggests not. Between February 2022 and March 2026, output in the energy-intensive branches fell by 15.2 percent while employment fell by 6.3 percent. Output per employee therefore fell by about 9.5 percent. In an industry with high fixed costs and long training times, firms hoard skilled labour through downturns in the expectation of recovery. But hoarding is costly, and it is sustainable only as long as the recovery remains plausible.

We define the labour overhang as the number of employees in excess of what current output would require at the productivity of February 2022. With 794,400 employees in March 2026 and a productivity gap of 9.5 percent, the overhang is about 75,000 jobs. This is an original measure of the latent adjustment that the survivors are carrying. It is also a measure of their fragility: if a new shock persuades firms that the recovery will not come, the adjustment of employment could be abrupt. The automotive industry offers a warning. Its employment fell to the lowest level since 2005 in 2026, with 42,300 jobs lost in a single year, after years in which output had fallen faster than headcount.

5.4 Interpreting survivor elasticity

Taken together, Sections 3 to 5 support a simple interpretation. The 2022 shock acted as a filter. It removed the capacity with the lowest gas price thresholds, quickly and permanently, which is why the 2022 elasticity was high and why output did not recover when prices fell. The capacity that remains has thresholds above the levels seen so far in 2026, which is why the short-run response this year has been small. But the survivors carry a labour overhang, face structural competition, and in chemicals their threshold is eroding. Their low elasticity is not a sign of strength; it is a sign that the next adjustment, when it comes, is more likely to be a step than a slope.

This has a direct implication for the winter. A gradual increase in gas prices, even to 85 or 100 EUR/MWh, is unlikely to cause large output losses in the energy-intensive branches. What could cause large losses is an event that pushes the six-month average price towards the survivors’ threshold, or that forces physical curtailment of gas supply regardless of price. Both depend on the state of German gas storage, to which we now turn.

6. Winter 2026/27: A Monte Carlo Model of Gas Storage

6.1 The starting point

German gas storage was 61.4 percent full on 1 November 2026, according to the latest AGSI data cited by the storage association. Since the storage regulation requires most facilities to be 80 percent full and six facilities, including Rehden, to be 45 percent full on that date, the effective national target was about 70 percent, which was missed by about 8.6 points. The working gas capacity implied by the government’s answer to a parliamentary question, 126.8 TWh at 51.2 percent on 26 August 2026, is about 248 TWh. The November fill therefore corresponds to about 152 TWh, consistent with the association’s estimate of about 150 TWh.

The path that led there is itself a measure of the shock. Storage started the 2025/26 winter at about 76 percent, fell to 37 percent by the end of a cold January and to 21 percent in early March, and stood at 22 percent on 1 April. It reached only 41 percent on 1 July, against 51 percent a year earlier, and 53.4 percent on 1 September, the lowest level for that date in 15 years of records. Injection averaged 0.7 TWh a day in June but only 0.4 TWh a day in July and August, when INES reported that injecting gas implied a loss of about 3 EUR/MWh. The government noted in September that about a quarter of German storage sites had already met their targets, but that three Bavarian facilities, Wolfersberg, Breitbrunn and Inzenham-West, would not.

The legal framework sets one further milestone that matters for our analysis: storage must still be at least 30 percent full on 1 February 2027. The fill-level rules expire on 31 March 2027 and will not be extended. A strategic gas reserve of 24 TWh, about 10 percent of capacity, is planned from 2027, too late for this winter.

6.2 Calibration

We model the net withdrawal from German storage in each month from November to March as the product of a normal-weather baseline and a weather factor, plus an optional supply stress term. Three public anchors calibrate the model.

Table 9. Calibration anchors of the storage model.

Anchor Source Value used
Normal winter (reference year 2016): 77% on 1 Nov to about 38% on 1 Apr INES scenario, summer 2026 Net withdrawal of 96.6 TWh, November to March
Observed winter 2025/26: 76% on 1 Nov to 37% at end-January AGSI, as reported Net withdrawal of 96.6 TWh, November to January
Very cold winter (reference year 2010) with 77% start: storage empty in early February INES scenario, September 2026 Validation point for the cold tail
Working gas capacity Bundestag document 21/7879 247.7 TWh
Start level, 1 Nov 2026 AGSI, as reported 61.4% (152.1 TWh)

Source: INES, AGSI and Bundestag, as cited; author's calculations.

The coincidence in the table is instructive. Last winter’s withdrawal between November and January alone equalled the withdrawal that INES assumes for a whole normal winter. Either last winter was unusually cold, which is partly true given the cold January, or the INES normal scenario assumes stronger imports than the market delivered. We resolve this by treating INES’s normal scenario as the median and last winter’s withdrawal as the 80th percentile of the November to January distribution. Monthly weather factors are lognormal, with a common log standard deviation of 0.59 that is calibrated to satisfy this condition, and are autocorrelated from month to month with a coefficient of 0.5. The baseline is distributed across the five months in proportions of 14, 22, 27, 23 and 14 percent.

The supply stress term represents events that reduce net imports independently of German weather: further LNG diversion to Asia while Qatari volumes are missing, Norwegian outages, or higher exports to neighbours whose own storage is low (the Netherlands stood at 45 percent at the end of August). We assume a 30 percent probability that such stress occurs during the winter, starting in a random month and adding 3 to 12 TWh to net withdrawals in every remaining month. Storage cannot fall below a technical floor of 5 percent of capacity; demand that cannot be met from storage above that floor is recorded as a shortfall.

As a validation, the model reproduces the INES normal path by construction, and a winter two standard deviations colder than normal with supply stress, starting at 77 percent, empties storage before 1 February, consistent with INES’s very cold scenario. We run 100,000 simulated winters.

6.3 Results

0102030405060701 Nov1 Dec1 Jan1 Feb1 Mar1 Apr% of capacity30% legal floor on 1 Feb20% stress lineMedian, 25 to 75 and 5 to 95 percentile bands
Figure 6. Monte Carlo storage paths for winter 2026/27 starting from 61.4 percent on 1 November: median and percentile bands. Source: author's simulation.

Table 10. Storage Monte Carlo results, baseline (100,000 draws).

Statistic Value
Median fill on 1 February 2027 34.2%
10th to 90th percentile on 1 February 12.0% to 46.6%
Probability below 30% on 1 February (legal floor) 37.9%
Probability below 20% on 1 February 18.4%
Probability below 10% on 1 February 8.6%
Median fill on 1 April 2027 16.0%
Probability that storage reaches the technical floor before April 29.6%
Mean shortfall across all simulations 13.2 TWh
Mean shortfall, conditional on a shortfall occurring 44.6 TWh

Source: author's simulation.

The median winter is uncomfortable but manageable: storage would be about 34 percent full on 1 February, above the legal floor, and about 16 percent full on 1 April. But the distribution is wide and skewed. In almost two simulated winters out of five, storage would be below the 30 percent floor on 1 February. In almost one in five it would be below 20 percent, a level at which withdrawal capacity begins to fall and price spikes become likely. In three out of ten, the system would reach the technical floor before the end of March and some demand would have to be met by other means: additional LNG at very high prices, voluntary demand reduction, or curtailment.

6.4 Four weather scenarios

For readers who prefer scenarios to probabilities, Table 11 reports four deterministic paths. The warm and very cold scenarios correspond to one standard deviation below and two standard deviations above normal; the third scenario replays the observed monthly withdrawals of winter 2025/26 from this year’s lower starting point.

Table 11. Storage paths under four weather scenarios, percent of working gas capacity. The very cold shortfall is a mechanical gap before any market response.

Scenario 1 Dec 1 Jan 1 Feb 1 Mar 1 Apr Withdrawal (TWh) Shortfall (TWh)
Warm (1 sd) 58.4 53.6 47.8 42.9 39.9 53 0
Normal (INES 2016) 55.9 47.4 36.8 27.9 22.4 97 0
Replay of winter 2025/26 51.7 38.8 22.4 6.4 7.4 134 0
Very cold (2 sd), supply stress 40.3 8.9 5.0 5.0 5.0 357 217

Source: author's calculations.

The replay scenario is the most useful single number in this section, because it requires no assumption about probabilities. If the coming winter is like the last one, German storage will be at 22 percent on 1 February, eight points below the legal floor, and close to empty in early March. Last winter ended without incident because it started at 76 percent. This one starts 15 points lower.

6.5 Sensitivity

Table 12. Sensitivity of storage risk to the starting level and model assumptions.

Variant Below 30% (1 Feb) Below 20% (1 Feb) Below 10% (1 Feb) Floor before April
Baseline (start 61.4%) 37.9% 18.4% 8.6% 29.6%
Start 70% (effective target) 20.4% 9.6% 4.5% 18.6%
Start 77% (INES feasible maximum) 12.1% 5.6% 2.7% 12.5%
No supply stress 34.7% 16.6% 7.7% 24.8%
High supply stress (50%, 5 to 20 TWh a month) 44.0% 22.7% 10.9% 39.2%
Winter 2025/26 at 70th weather percentile 44.3% 28.3% 18.2% 39.7%
Winter 2025/26 at 90th weather percentile 30.4% 9.1% 2.3% 18.9%
No month-to-month weather correlation 41.5% 18.2% 7.5% 31.1%
Strong weather correlation (0.8) 36.2% 18.5% 9.3% 28.9%

Source: author's simulation.

The starting level is by far the most important parameter. Each of the 15.6 points between the actual November fill and the 77 percent that INES considered technically feasible would have reduced the probability of breaching the 30 percent floor by roughly 1.7 points. The weather calibration matters for the tail: if last winter is considered less exceptional (70th rather than 80th percentile), the probability of falling below 20 percent rises to 28 percent; if it is considered more exceptional, it falls to 9 percent. No plausible variant brings the probability of breaching the legal floor below 10 percent.

7. From Gas to Output and GDP

7.1 Three channels

We now translate the storage distribution into energy-intensive output and GDP. The model distinguishes three channels, each grounded in the estimates of the previous sections.

The price channel. For each simulated winter, the gas price is drawn around the November contract price of 85 EUR/MWh with a dispersion of 15 percent, and is then raised by a scarcity premium that doubles the price for every 10 points by which storage on 1 February falls below 30 percent, capped at 2.75 times the base, which corresponds to the 2022 peak. The temporary output effect over the five winter months is the survivor elasticity (drawn between minus 0.04 and zero, mode minus 0.02, consistent with Tables 4 and 5) times the log price increase relative to the pre-war base of 31 EUR/MWh.

The quantity channel. When storage reaches its technical floor, the mechanical shortfall must be closed. We assume that 30 to 70 percent of it is closed by market responses outside the energy-intensive branches (additional LNG, fuel switching in power generation, voluntary reduction by other industries and commerce) and that the energy-intensive branches bear 50 to 100 percent of the remainder, up to half of their winter gas use. These branches used about 200 TWh of natural gas a year, derived from their 2024 energy consumption of about 2,738 PJ and the 26.3 percent share of natural gas, or about 22 percent of Germany’s 2025 gas consumption of 910 TWh. Each percent of gas curtailed reduces output by 0.6 to 1.0 percent; the central value of 0.85 reflects the 2022 and 2023 experience, when the energy consumption of these branches fell by 9.6 and 8.4 percent.

The permanent channel. If the six-month average price over November to April exceeds the survivors’ ratchet threshold, drawn between 120 and 165 EUR/MWh in line with Table 6, output falls permanently with a ratchet elasticity drawn between 0.15 and 0.29, the range estimated for metals and chemicals.

Losses are expressed as a percentage of annual energy-intensive output. GDP effects use the share of the energy-intensive branches in gross value added, 3.4 percent in 2021 according to ifo, scaled down by 17 percent to reflect their lower output, and a supply-chain multiplier drawn between 1.0 and 2.0 (mode 1.4).

7.2 Results

Table 13. Energy-intensive output and GDP at risk, winter 2026/27 (100,000 draws).

Statistic Energy-intensive output (% of annual) GDP (percentage points)
Median minus 1.1 minus 0.05
Mean minus 4.4 minus 0.18
5th percentile minus 21.9 minus 0.90
1st percentile minus 29.6 minus 1.27
Mean of price channel minus 1.1
Mean of quantity channel minus 1.8
Mean of permanent channel minus 1.5
Probability of new permanent losses 21%
Conditional on storage below 20% on 1 Feb minus 17.0 minus 0.70

Source: author's simulation.

-30-25-20-15-10-50P5: minus 21.9%Median: minus 1.1%Output effect, % of annual energy-intensive output (bar height on log scale)
Figure 7. Distribution of the energy-intensive output effect in winter 2026/27, percent of annual output; bar heights on a logarithmic scale. Source: author's simulation.

The distribution has a narrow centre and a long left tail. In the typical winter, the survivors absorb the price increase with a loss of about 1 percent of annual output, equivalent to about 0.05 percentage points of GDP. That is the payoff of survivor elasticity. But in about one simulated winter in five, a combination of cold weather, supply stress and low storage pushes prices above the survivors’ threshold or forces physical curtailment, and the losses become an order of magnitude larger. In the worst 5 percent of outcomes, energy-intensive output loses more than a fifth of a year’s production, comparable to the 2022 shock, and GDP is reduced by close to a point.

The mean of 4.4 percent is driven almost entirely by the tail. The permanent channel, which is zero in the median winter, accounts for a third of the expected loss. This is the quantitative form of our thesis: the 2026 shock is less dangerous than 2022 at the margin, but the danger that remains is concentrated in discontinuous, irreversible events.

Table 14. Output effects by weather scenario, percent of annual energy-intensive output, central parameter values.

Scenario Storage 1 Feb (%) Price (EUR/MWh) Price ch. Quantity ch. Permanent ch. Total GDP (pp)
Warm 47.8 85 minus 0.9 0.0 0.0 minus 0.9 minus 0.03
Normal 36.8 85 minus 0.9 0.0 0.0 minus 0.9 minus 0.03
Replay 2025/26 22.4 144 minus 1.3 0.0 0.0 minus 1.3 minus 0.05
Very cold, stress 5.0 235 minus 1.7 minus 17.7 minus 8.5 minus 27.9 minus 1.10

Source: author's calculations.

The scenario table shows how sharply the risk is concentrated. Even the replay of last winter, with storage eight points below the legal floor on 1 February and a winter price of about 144 EUR/MWh, leaves the survivors largely intact: the six-month average would remain just below the central threshold. Only the very cold scenario breaks through, and when it does, the quantity and permanent channels together are fifteen times larger than the price channel.

7.3 The survivor dividend

What if the industry had kept its 2022 elasticity? To answer this counterfactual, we rerun the model with the temporary elasticity set at the 2022 level of about minus 0.17. The expected output loss rises from 4.4 percent to 12.2 percent, and the expected GDP cost from 0.18 to 0.51 percentage points. The difference, about 0.33 percentage points of GDP, is what we call the survivor dividend: the protection that the 2022 shakeout, painful as it was, now provides against a proportional price shock.

The dividend is real, but it should not be celebrated. It was paid for in advance by the permanent loss of about a sixth of energy-intensive capacity and more than 50,000 jobs. And it protects only against the price channel. Against the quantity and permanent channels, which dominate the tail, the survivors are, if anything, more exposed than in 2022, because storage is lower, the threshold is eroding and the labour overhang is larger.

7.4 Sensitivity

Table 15. Sensitivity of output and GDP at risk.

Variant Mean output (%) P5 output (%) Mean GDP (pp) P(permanent loss)
Baseline minus 4.4 minus 21.9 minus 0.18 21%
Start 70% on 1 November minus 2.8 minus 17.0 minus 0.12 11%
Start 77% on 1 November minus 2.1 minus 11.3 minus 0.09 7%
No supply stress minus 3.9 minus 20.5 minus 0.16 19%
Winter 2025/26 at 90th weather percentile minus 2.4 minus 13.1 minus 0.10 12%
Winter 2025/26 at 70th weather percentile minus 7.0 minus 27.7 minus 0.29 30%
Ratchet fixed at 2022 peak (164) minus 3.7 minus 18.7 minus 0.15 17%
Ratchet eroded to 110 minus 5.7 minus 27.4 minus 0.24 31%
No market offset of shortfalls minus 5.4 minus 26.5 minus 0.22 21%
Temporary elasticity at 2022 level minus 12.2 minus 34.8 minus 0.51 21%

Source: author's simulation.

Two parameters dominate. The first is the starting level of storage: filling to 77 percent would have halved the expected output loss and cut the probability of new permanent closures by two thirds. The second is the position of the ratchet: if the survivors’ threshold has eroded to 110 EUR/MWh, the probability of permanent losses rises to 31 percent. Both are, to a degree, under the control of policy.

-15-10-5050100150200250Permanent loss, %Jun to Nov 2026: 61Ratchet at 2022 peak (164)Eroded 5% a year (136)Further erosion (110)Six-month average TTF price, EUR/MWh
Figure 8. Permanent output loss as a function of the six-month average gas price for three positions of the ratchet threshold. Source: author's calculations based on Table 6.

8. Policy Implications

The analysis does not suggest that German energy-intensive industry is about to repeat 2022. It suggests that the nature of the risk has changed, and that policy should change with it. Five implications follow.

First, protect the gas balance before protecting the price. Because the short-run output elasticity of the survivors is low, broad subsidies that lower the gas or power price for energy-intensive firms would buy little additional output this winter, at considerable fiscal cost. The same money spent on securing additional LNG cargoes for January to March, on contracting interruptible demand in advance, or on tenders that keep gas in storage until February would act on the channel that dominates the tail of the distribution. The government’s position in September 2026 that no shortage was expected and that fill targets need not change was defensible on the median, but not on the distribution.

Second, make curtailment predictable. The quantity channel causes damage partly because it is disorderly. Firms that do not know whether and when they will be curtailed cannot plan production, honour contracts or protect equipment. An auction for voluntary industrial demand reduction, priced in advance and activated when storage falls below a pre-announced trigger such as 25 percent, would turn a potential shock into a priced option. It would also identify, before the crisis, which processes can be interrupted at least cost.

Third, treat the ratchet threshold as a policy variable. Our estimates suggest that the survivors’ threshold for permanent closure is falling, by about 5 percent a year in chemicals. Measures that raise it, such as faster permitting for electrification, long-term contracts for low-carbon energy, and stable treatment of carbon costs for exposed sectors, have value precisely because they reduce the probability of the irreversible step change. Measures that only lower the current price do not move the threshold.

Fourth, fix storage policy for the years after 2027. The fill-level rules expire at the end of March 2027, and the strategic reserve of 24 TWh will only be built up gradually from 2027. Our model implies that each point of storage on 1 November reduces the probability of breaching the February floor by almost two points. A storage regime that relies on market incentives alone failed in 2026 because the price structure made injection unprofitable. A permanent instrument, whether a strategic reserve, a fill obligation with a cost-sharing mechanism, or contracted storage capacity, is cheaper than the expected output loss it prevents.

Fifth, monitor the survivors, not only the aggregate. The aggregate production index of the energy-intensive branches is distorted by mineral oil processing, which benefits from the same war that harms chemicals. Policymakers should watch a small set of leading indicators: relative chemical output, the six-month average gas price relative to the ratchet threshold, the labour overhang, and the storage path relative to the legal floor. Annex C proposes such a dashboard.

9. Conclusion

Germany’s second energy shock in four years is hitting an industry that has already been reshaped by the first. The 2022 shock removed the most gas-sensitive capacity in a matter of months, and our survivor decomposition shows that almost none of it came back when prices fell. A hysteresis model, estimated on the period before the Iran war, shows that the 2022 loss was a ratchet: output fell with the highest sustained price reached, not with the current price. The same model, applied out of sample, predicts the small response observed in 2026, because prices have so far remained well below the 2022 peak.

The resulting survivor elasticity is low. Chemical output responds to gas prices four to five times less than in 2022, and metals and intermediate goods barely respond at all. In a normal winter, the energy-intensive branches will absorb the shock with output losses of about 1 percent. This is the survivor dividend, worth about a third of a percentage point of GDP compared with a world in which the industry still had its 2022 elasticity.

But low elasticity is not resilience. The survivors carry a labour overhang of about 75,000 jobs, their threshold for permanent closure is eroding, and they face the winter with German storage at a record low. Our Monte Carlo model gives a probability of almost two in five that storage will breach the legal floor on 1 February, and of almost one in five that it will fall below 20 percent. In those winters, the price and quantity shocks become large enough to break through the ratchet, and the losses resemble 2022.

The original findings of this paper can be summarised in five numbers: a 2026 elasticity of about minus 0.05 against minus 0.19 in 2022; a recovery of only 8 percent of the 2022 loss by 2025; a ratchet threshold of 136 to 164 EUR/MWh against a current six-month average of 61; a 38 percent probability of breaching the storage floor; and an expected output loss of 4.4 percent with a 5 percent tail beyond 22 percent. Together they describe an industry that is harder to bend and easier to break. Policy should be designed for that industry, not for the one that existed in 2022.

About this series. This paper is the second in a ten-part series on the German economy in 2026 published by Global Strategy Files. The other papers examine the absorption of the EUR 500 billion infrastructure fund, the irreversibility threshold of German chemicals, a county-level map of the China shock in the automotive industry, rearmament and labour reallocation, the end of Bund scarcity, pension anxiety and household saving, the municipal fiscal crisis, the insolvency wave, and the Beveridge curve paradox of a shrinking workforce.

Annex A. Methodology

Data. Monthly seasonally adjusted production indices (2021 = 100) for chemicals and pharmaceuticals (WZ 20 and 21), metals and metal products (WZ 24 and 25), intermediate goods producers and total industry, September 2021 to September 2026, are taken from the Bundesbank’s Saisonbereinigte Wirtschaftszahlen, tables III.1a to III.1c, published on 6 November 2026; values for July to September 2026 are provisional. Monthly changes in the Destatis production index of the five energy-intensive branches are taken from Destatis press releases and their summaries. Gas prices are the monthly TTF index of the European Energy Exchange applied to each delivery month, as published by Protergia, January 2022 to November 2026. Storage data are taken from AGSI as reported by INES, the press and the Bundestag.

Reconstruction of the energy-intensive aggregate. The level in March 2026 (83.8) is taken from a Destatis-based report; the level in February 2022 is implied by the 15.2 percent decline to March 2026. Levels for January to September 2026 are chained from first-release monthly changes; levels for January to September 2025 are derived from the 2026 year-on-year changes. December 2022 is chained from the September 2022 year-on-year decline, assuming that the seasonally adjusted index in September 2021 was at its 2021 average. Annual levels for 2023 and 2024 are derived from the reported annual changes. Because first releases are revised, the reconstruction carries an uncertainty of about one index point.

Episode elasticities. E = [log Y(window) minus log Y(base)] / [log P(window) minus log P(base)], with and without subtracting the corresponding change in total industry. Standard errors assume that monthly deviations of relative output around their local mean have the standard deviation observed in 2024 and 2025.

Hysteresis model. r(t) = a + b log M(t) + c log P(t) + e(t), where r is 100 times the log of branch output relative to total industry, P is the six-month (or three-month) moving average of the TTF index from January 2022, and M(t) = max[P(t), M(t minus 1) x (1 minus d)^(1/12)]. Estimated by OLS on January 2022 to February 2026 with Newey-West standard errors (6 lags); d chosen by grid search between 0 and 0.30. The ratchet share of the 2022 loss is b x [log M(Dec 2022) minus log M(Jan 2022)] divided by the actual change in r.

Storage model. Monthly net withdrawal W(m) = B(m) x exp(s x z(m)) + stress(m), with B the INES normal profile (96.6 TWh in total, distributed 14, 22, 27, 23 and 14 percent), z an AR(1) standard normal process with coefficient 0.5, and s = 0.59 calibrated so that the November to January withdrawal of 2025/26 (96.6 TWh) is the 80th percentile. Stress occurs with probability 0.3, starts in a random month and adds a uniform 3 to 12 TWh a month thereafter. Capacity 247.7 TWh, floor 5 percent, 100,000 draws, fixed seed.

Output at risk. Price: P = 85.3 x exp(N(0, 0.15)) x min[2^((30 minus S_Feb)/10), 2.75] when S_Feb is below 30. Price channel: e_T x log(P/30.6) x 5/12, e_T triangular (minus 0.04, minus 0.02, 0). Quantity channel: curtailment = min[shortfall x (1 minus phi) x u, 0.5 x 200 x 5/12], phi uniform (0.3, 0.7), u uniform (0.5, 1); output loss = curtailment / 200 x k, k triangular (0.6, 0.85, 1.0). Permanent channel: b_max x log(max[P6 / threshold, 1]), b_max triangular (0.15, 0.22, 0.29), threshold uniform (120, 165), P6 the November to April average price with April at the base price. GDP: output effect x 0.034 x 0.83 x multiplier, multiplier triangular (1.0, 1.4, 2.0).

Limitations. The estimation sample is short and dominated by one large episode, so the ratchet and temporary coefficients are identified mainly by the contrast between 2022 and 2023 to 2025. The Bundesbank chemical series includes pharmaceuticals. The month-ahead index is a proxy for the effective gas cost of firms, many of which hedge for longer horizons. The storage model is mechanical: it does not model the price-induced increase in LNG imports or the reduction in demand that would occur before storage reached its floor, although the output model allows 30 to 70 percent of any shortfall to be met in this way. The weather calibration rests on treating last winter as an 80th percentile outcome; Table 12 shows that this choice matters for the tail. We were unable to verify within this study the retail and industrial electricity price data and the terms of the industrial power price measures under discussion, so the paper does not use them; the electricity channel is therefore absent from the output model, which makes the estimates conservative. All results should be read as calibrated scenarios, not forecasts.

Annex B. Data Tables

Table B1. Data used in this paper.

Indicator Value Date or period Source
Energy-intensive production, change since Feb 2022 minus 15.2% March 2026 Destatis
Industry production, change since Feb 2022 minus 9.5% March 2026 Destatis
Energy-intensive production, Q3 vs Q2 2026 minus 3.5% Q3 2026 Destatis
Energy-intensive production, y/y minus 1.9% September 2026 Destatis
Energy-intensive production, annual change plus 0.6% / minus 2.6% 2024 / 2025 Destatis
Energy-intensive production, y/y minus 9.7% / minus 12.6% / minus 12.9% / minus 19.6% Sep / Oct / Nov / Dec 2022 Destatis
Energy-intensive employment 794,400 (from 847,700) March 2026 Destatis
Share of industrial energy use 77% (2021); 75.6% (2024) 2021, 2024 Destatis
Share of industrial gross value added 17% 2021 Destatis
Real gross value added EUR 98 bn (3.4% of the economy) 2021 ifo
Energy carriers: gas / oil / coal / electricity 26.3% / 21.0% / 18.1% / 15.2% 2024 Destatis
Energy consumption of energy-intensive branches about 2,738 PJ 2024 Destatis
Change in energy consumption minus 9.6% / minus 8.4% / plus 3.2% 2022 / 2023 / 2024 Destatis
German gas consumption 910 TWh (plus 6.9%) 2025 Cleanthinking
Import shares: Norway / Netherlands / Belgium / LNG 44% / 24% / 21% / 10.3% 2025 Cleanthinking
Storage 61.4% (about 150 TWh) 1 November 2026 AGSI via INES
Storage 126.8 TWh (51.2%) 26 August 2026 Bundestag 21/7879
Storage 76% / 37% / 21% / 22% 1 Nov 2025 / end Jan / early Mar / 1 Apr 2026 Cleanthinking
Storage 41% / 53.4% 1 Jul / 1 Sep 2026 Cleanthinking
Legal floor 30% 1 February 2027 Gas storage regulation
Strategic gas reserve (planned) 24 TWh from 2027 INES, press
German gas imports, value EUR 15 bn (EUR 14.6 bn at 2025 prices) H1 2026 Bundestag 21/7879
Ras Laffan capacity out of service 17% from 18 March 2026 Cleanthinking

Annex C. A Dashboard for Survivor Risk

We propose six indicators, all published at least monthly, to track whether the survivors are approaching their breaking point.

Table C1. Survivor-risk dashboard. Thresholds: author's proposal.

Indicator Publication Warning threshold Current reading
Storage relative to legal path Daily, AGSI More than 5 points below a path reaching 30% on 1 Feb 61.4% on 1 Nov, 8.6 points below target: warning
Six-month average TTF vs ratchet threshold Monthly, EEX Above 75% of the threshold (about 100 EUR/MWh) 61 EUR/MWh: watch
Relative chemical output (vs industry) Monthly, Bundesbank More than 3 points below the 2026 base for 3 months minus 1.9 (Q3): watch
Energy-intensive production excluding mineral oil Monthly, Destatis (on request) Below the December 2022 trough Aggregate at trough level: warning
Labour overhang Quarterly, Destatis Rising while output falls about 75,000: watch
Winter-summer price spread Daily, EEX Summer above winter (injection loss-making) Injection loss reported in 2026: warning

References

Bundesbank (2026). Saisonbereinigte Wirtschaftszahlen, Tables III.1a, III.1b and III.1c: Produktion im Produzierenden Gewerbe. Frankfurt am Main, 6 November 2026.

Deutscher Bundestag (2026). Antwort der Bundesregierung: Stand der Gasversorgung und Einspeicherung im Sommer 2026. Drucksache 21/7879, 3 September 2026.

Deutscher Bundestag (2026). Kleine Anfrage: Versorgungssicherheit Deutschlands mit Erdgas im Winter 2026/2027. Drucksache 21/8255, 29 September 2026.

Destatis (2022 to 2026). Press releases on production in industry, including the energy-intensive industrial branches: reporting months September, October and December 2022; January 2024; December 2024; December 2025; January to September 2026.

Destatis (2025). Energy consumption in industry 2024, press release of 14 November 2025, as published by Presseportal.

Destatis (2026). Production in energy-intensive industries since February 2022, May 2026, as reported by Produktion, onvista, Clean Energy Wire and Finanzmarktwelt.

Cleanthinking (2026). Historisch niedrige Fuellstaende: Reicht das Gas fuer den Winter 2026/27? 8 September 2026.

Schaller, L. and Schasching, M. (2024). Energieintensive Industrie in Deutschland. ifo Schnelldienst 6/2024.

INES, Initiative Energien Speichern (2026). Scenarios for winter 2026/27, as reported by ZfK, ZDFheute, Berliner Zeitung, Stuttgarter Zeitung, infranken and Political.org, July to November 2026.

Protergia (2026). TTF prices per month, based on the EEX monthly index, accessed November 2026.

Statista (2026). Produktionsindizes energieintensiver Industriezweige in Deutschland, 18 May 2026.

FXStreet (2026). European natural gas surges 85 percent on Qatar LNG shock and Hormuz disruptions, 3 March 2026.

Vogel, L., Neumann, M. and Linz, S. (2023). Calculation and development of the new production index for energy-intensive industrial branches. WISTA 2/2023, Destatis.

Driss EFFINA, PhD
Independent Researcher, Global Strategy Files

Free preview
The Q4 2026 Global Outlook

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.