The price elasticity of natural gas demand of small consumers in Germany during the energy crisis 2022

Since mid-2021, Europe has been navigating an energy crisis. In the fall of 2021, Russian exports of natural gas to Europe dropped to a third of previous years’ levels. In combination with high LNG demand in Asia and low European production, this led to low European storage levels and substantial price increases (Fulwood, 2022). Following the Russian invasion of Ukraine in February 2022, European import flows from Russia declined even further. Eventually, the Nord Stream pipeline was completely shut down in summer 2022. During this period, the price of natural gas reached unprecedented levels along with the fear of shortages.

When the war broke out, the European natural gas market was highly reliant on natural gas supplies from Russia. Until the second half of 2021, Russia delivered around 50% of EU imports. In May 2022, the EU launched the REPowerEU strategy aiming to diversify supplies and ensure ”affordable, secure and sustainable energy for Europe”. In addition, the EU member states agreed to a specific savings target in June 2022, committing to reduce their natural gas demand by 15% compared to the average consumption in the past five years.

In a longer perspective, the climate crisis requires a rapid reduction in emissions and a shift away from fossil fuels including natural gas. The European Green Deal has set a goal of achieving net-zero emissions by 2050, which requires a major transition in the European energy system. During a transition phase, natural gas capacities could serve as a backup in a highly renewable electricity system to meet the goal of ”secure energy supply” highlighted in REPowerEU. Therefore, the concern of security of natural gas supply might not be over as Europe recovers from the 2022 market disruption.

Natural gas makes up around a quarter of Germany’s final energy consumption (IEA, 2022). Figure 1a shows that the residential sector represents the largest share of German natural gas consumption, followed by industry and the commercial sector. For households, Fig. 1 illustrates that space and water heating are the major end uses of natural gas. Almost half of German homes use natural gas for heating, and despite energy transition targets, 70% of newly installed heating systems are based on natural gas (BMWK, 2022).

Full size image

Individual consumption and price level at the household level is not available. Instead, we analyze aggregated, total daily consumption data from all households and small and medium-sized enterprises in Germany (also see Appendix A for more details). This group, referred to as small consumers, reports their yearly natural gas consumption by reading an analog gas meter. Their (monthly) bills are calculated ex-post based on their yearly consumption. Gas distribution companies operate the gas grid on a daily basis by forecasting a Standard Load Profile (SLP) for small consumers. This SLP takes into account the building stock’s characteristics, the types of heat technologies and changes in weather. If the actual metered consumption during a 12-month-period exceeds the estimated amount, consumers need to pay for the additional usage. Conversely, if they have used less than predicted, they are refunded the difference (Trading Hub Europe, n.d.).

The available literature on the price elasticities of demand for natural gas is limited and exhibits significant variations in the reported estimates. Most studies investigate the responsiveness of economy-wide natural gas demand, partly because there is often a lack of more disaggregrated data for specific consumer groups. The seminal studies of Balestra and Nerlove (1966) and Pindyck (1979) lay the foundation for later research on natural gas demand, estimating long-run price elasticities of -0.63 and -1.7, respectively. More recent studies find the demand of natural gas to be less elastic. For instance, Asche et al. (2012) analyze residential natural gas demand in 12 European countries from 1978 to 2002, finding a price elasticity in the range -0.10 to 0 in the short run and -0.60 to 0 in the long run, using a shrinkage estimator. Similarly, in their meta-analysis of energy elasticities, Labandeira et al. (2017) find an average short run elasticity of -0.184 and a long run elasticity of -0.568 for natural gas.

Over the last decade, error-correction-type models have become more prominent in the literature, addressing the issue of non-stationarity. Among these, (Bernstein & Madlener, 2011) investigate the price elasticity of demand in 12 OECD countries, estimating a short-run elasticity of -0.24 and a long-run elasticity of -0.51 with the ARDL model. Erias and Iglesias (2022) is among the most recent studies, and the only study to our knowledge to analyze price elasticities of natural gas demand in Europe using wholesale prices and a daily frequency. They report rather diverging estimates in the range -2.2 to 0 for the short-run elasticity and -0.94 to -4.3 for the long-run elasticities.

Another direction of recent studies makes use of microdata to analyze natural gas demand. Rubin and Auffhammer (2024); Filippini and Kumar (2021), and Alberini et al. (2020) address endogenous variables by employing an lagged variable approach. Rubin and Auffhammer (2024) exploit spatial discontinuity between two natural gas utilities in California and find a price elasticity of demand of -0.19 to -0.15. Filippini and Kumar (2021) estimate the price elasticity of -0.73 for Swiss residential gas demand for the period from 2010 to 2014. Alberini et al. (2020) find a price elasticity of demand of -0.16 in the short run analyzing a period of large price variations in Ukraine.

Most of the estimates found in the literature appear not be in line with the developments observed during the energy crisis following the invasion of Russia in Ukraine when natural gas demand decreased substantially. Hence, we provide an updated estimate of the price elasticity of demand in a period of surging prices. Consumer contract (retail) prices, as calculated by the Verivox retail price index, increased by 143% in 2022 compared to the yearly average between 2018 and 2021 (see Section “Data” for more details). In the same period, we observe a reduction in yearly demand of 11%, not adjusted for weather. Even if we assume that the demand reduction was driven by price in its entirety, an estimate of price elasticity of demand below \(\frac{-11\%}{143\%} = -0.077\) appears to be quite substantial.

First analyses of the 2022 energy crisis have pointed out a considerable reduction in German natural gas consumption. Roth and Schmidt (2023) investigate the demand reduction from small consumers, adjusted for weather effects, and find a total demand reduction from small consumers of 23TWh during the period September-December 2022. Ruhnau et al. (2023) also find a weather-adjusted reduction in the same magnitude and discuss the drivers qualitatively. We expand their analyses by quantifying the components of the demand decline.

German small consumers typically have fixed energy price contracts and are not directly exposed to wholesale prices. However, during periods of tight supply, a demand-side response is needed to avoid shortages. The wholesale market efficiently matches supply and demand for natural gas, with the market price reflecting real-time information on supply levels, storage inventories, weather patterns and geopolitical events. We aim to understand how small consumers respond to market signals by estimating the wholesale price elasticity of demand. This is an extension of the existing literature, where this relationship has not been investigated in detail yet. In addition, we estimate the retail price elasticity of demand, representing the price that appears on the bills of German consumers.

The German single market place Trading Hub Europe (THE) publishes aggregate consumption data on natural gas. They divide the natural gas customers into two groups. Customers with daily metering (RLM) and customers with manual metering whose daily consumption is estimated in a standard load profile (SLP). SLP customers primarily consist of households and small and medium-sized enterprises who use natural gas for space heating, water heating and cooking.

We use the actual daily consumption of all SLP customers in Germany, as it is reported ex-post with a few days lag after realization. SLP demand as calculated by the gas grid operators as the residual load, more precisely as ”the sum of all measured inputs minus all measured offtakes in a network” (Trading Hub Europe, n.d.).¹

Natural gas consumption is characterized by a strong seasonal trend in both the mean and the variance. The need for heating, and thereby the consumption of natural gas, is highest during the winter, while the typical summer consumption for water heating and cooking is lower and more stable (Fig. 2).

The wholesale price signal was highly mediatized in most of the analysed period and became almost equally well-known by consumers than their individual contract price. To evaluate the effects of wholesale market prices, we use closing day-ahead prices from the European Energy Exchange (EEX, 2023) in Germany, extracted from Bloomberg. The day-ahead series represent the prices for physical delivery of natural gas on the following working day. To avoid discontinuity during weekends and holidays, we incorporate the corresponding weekend-ahead forward price to the time series, also collected from Bloomberg.

Figure 3a shows the wholesale price series from 2018 to 2023. After a supply disruption from the Dutch Groningen field caused a short-run price rally in 2018, prices were relatively low and stable for a long period. In 2021, wholesale prices started to rise, reaching the highest peak in September 2022 following the Nord Stream 2 gas leaks. In the following months, prices fell distinctly before rising to a new spike at the end of the year.

Full size image

Data on retail prices are not available on a daily frequency. Moreover, in Germany, each household has an individual, unregulated contract price which is modified irregularly, sometimes less than once per year. The contract information is private and not available. German retailers have the right to adjust prices in existing contracts and from occasional anecdotal media reporting it is known that retail prices, also in existing contracts, are adjusted in response to high wholesale prices. Hence, we are confident that the price changes in new contracts follow the same (or at least a similar) pattern than the price changes in existing retail contracts. In the absence of precise information of prices paid by each household, we use a monthly index representing the end use price, the Verivox price index for natural gas ((Verivox GmbH, 2023), Fig. 3b). The index is only available at monthly frequency and not daily.

Through its price comparison website, Verivox enables consumers to evaluate energy tariffs. The Verivox retail price index represents the unit cost of natural gas for household end use, calculated based on an assumed annual consumption of 20,000kWh. This index incorporates both the variable and fixed price components of new contracts available on the Verivox website. The contributions of individual suppliers to the price index are weighted based on the number of households they serve within the supply area. Unlike the wholesale price series, the Verivox price index also encompasses all applicable taxes and fees. Both the retail prices and the wholesale price series have been deflated to the 2015-reference of the seasonally adjusted CPI index by OECD (2023).

To account for the impact of weather on natural gas demand, we gather a range of daily indicators from 587 weather stations operated by the German Meteorological Service (Deutscher Wetterdienst, 2023). From the temperature parameters, we engineer a Heating Degree Days (HDD) variable to establish a linear relationship between outdoor temperature and heating demand. We follow the approach used by Kendon et al. (2022) who for the UK Meteorological Office engineer a HDD variable based on temperature using daily mean, maximum and minimum values.

Full size image

Careful aggregation of the weather station data is crucial for obtaining reliable estimates, as weather is a predominant factor affecting heating needs.² In energy demand analysis, it is common practice to use population-weighted weather variables. However, population weighting ignores the spatial heterogeneity of natural gas dependence and dwelling sizes.³ Relying solely on population-based weighting would, therefore, lead to an under-representation of regions with a high dependence on natural gas.

To address these disparities, we incorporate detailed data from the German micro census (Destatis, 2019), which provides information on living area per capita and the share of natural gas-heated dwellings across 38 statistical regions. By multiplying these factors in a 1 km \(\times \) 1 km population density grid from Destatis (2022), we create a fine-meshed raster with the estimated natural gas-heated living space across Germany to use as aggregation weights (Fig. 4c).

In each grid cell i, we estimate the weather parameter value using Inverse Distance Weighting (IDW), assigning higher weights to closer weather stations, based on the assumption of spatial autocorrelation in weather parameters.⁴ Figure 4b illustrates this approach with heating degree days (HDD).⁵ The daily weighted average is then calculated using the interpolated weather parameters in each cell, with the estimated natural gas-heated living space within the cells as weights. Figure 4d shows the resulting weighted HDDs on a sample day with warm weather in the North-West.

We extract data from Google Trends to quantify the public awareness of the energy crisis. As Ruhnau et al. (2023) argue, the reduction in natural gas consumption in response to the crisis ”ould be driven by rising prices, anticipated future price increases, media attention on energy-related subjects, awareness of energy issues and conservation options, or, in the case of households, ethical considerations following the Russian invasion of Ukraine”. By disentangling the awareness effect from the price effect, we can also provide more precise estimates of the price elasticity of demand.

Full size image

We examine the relative frequency of Google queries from February 2022 to February 2023, following the Russian invasion of Ukraine. First, we identify the top 25 trending queries within the ”Energie” (energy) and ”Erdgas” (natural gas) topics. To avoid colinearity, we select the highest-ranking search term from each group of related queries, excluding those directly related to natural gas prices.

Out of the initial 50 candidates, we select the ”Energiekrise” (energy crisis) query as our indicator. This is both the most popular query, and the one with the lowest correlation to price. This last property is key, as we want to separate the price effect from other behavioral effects. We calculate a rolling mean over the past month to capture the sustained awareness beyond the specific query date.

Following the ad hoc approach to energy demand estimation proposed by Houthakker and Taylor (1970), we assume that natural gas demand from small consumers \(q_t\) at time t in equilibrium can be expressed as a linear function of k exogenous variables \(x_{i, t}\), \(i = 1, 2, ..., k\).Here, the intercept \(\mu \) and variable coefficients \({\beta _i}\) represent the equilibrium parameters to be estimated.

To unveil the relationship in Eq. 1, we estimate three distinct models: a daily model using wholesale prices as the price variable, and two monthly models to compare the demand response to wholesale prices to retail prices. We make use of the forward stepwise algorithm with the Akaike Information Criteria (AIC) to guide the variable selection. The descriptive statistics of the selected variables for either frequency are given in Appendix A.

Our models have a linear functional form, in contrast to the majority of the literature using log-log and log-linear specifications in modelling natural gas demand (i.e. Erias and Iglesias (2022); Asche et al. (2012) and Bernstein and Madlener (2011)). All these studies report weather as one of the most important determinants of natural gas demand. They use HDD to control for weather effects, despite the fact that HDD is designed to have a linear relationship with energy consumption for space heating.

Figure 5 unmistakably demonstrates this relationship. However, the figure also reveals that applying log-transformations to either the HDD or demand variable disrupts the original linear correlation. This insight presents a matter of concern, especially in light of the common practice in the literature of modeling log-transformed demand with HDD.⁶

The popularity of log-log specifications could be explained by the convenient feature that the coefficients can be interpreted as elasticities directly. In contrast, the coefficient estimates in our models represent the unit changes of natural gas demand q for unit changes in the explanatory variables. We can then interpret the price estimate as the slope of the demand curve. To obtain the corresponding (arc) elasticity of demand, \(\eta _p\), we calculate⁷where \(\frac{\partial {q}}{\partial {p}}\) is the coefficient estimate of price p and \(\frac{\bar{p}}{\bar{q}}\) is the ratio of the mean estimates of price and demand.

The standard error for \(\eta _p\), the product of a non-linear function, is calculated using the statistical error propagation function.

Sub-annual energy and weather data are characterized by the inherent seasonality stemming from the cyclical nature of Earth’s orbit.⁸ The standard approach to account for this would be to estimate the deterministic seasonality component \(S_t\) with T seasonal dummies \(D_s\) for period length T:Estimating seasonal dummies on a high frequency with a low number of periods can yield imprecise estimates, as the variances of the dummy coefficients are inversely proportional to the number of observations in each season s. The sensitivity for noise is apparent in the blue line in Fig. 6, where a set of \(T = 365\) day-of-year dummies is fitted to the daily demand series with only five whole periods from 2018-2023.

Full size image

Full size image

To separate the dominant seasonal patterns from noise and other effects, we turn to spectral analysis of the time series. Using discrete Fourier expansions, Eq. 3 can be represented with pairs of sines and cosines:where \(f_j = \frac{j}{T}\) is the Fourier frequency for harmonic \(j = 1, 2, ..., H\) and \(A_j\) and \(B_j\) are the respective amplitudes. The maximum number of harmonics is \(H = \frac{T}{2} - 1\) when T is even and \(H = {\left\lfloor {\frac{T}{2}}\right\rfloor }\) when T is odd. The components in Eq. 4 can be estimated on the data with ordinary least squares (OLS) and yield identical fitted values as the dummy components in Eq. 3. We can then evaluate the relative strength of each Fourier frequency by computing the periodogram:where \(\hat{A_j}\) and \(\hat{B_j}\) are the OLS estimates of the amplitudes in Eq. 4.

The periodogram for the daily data (Fig. 7) is calculated with a yearly periodicity of \(T = 365.24\) days. For natural gas demand and weather variables, we see a clear spike at the fundamental frequency \(f_1 = \frac{1}{T}\). This indicates a dominating annual seasonal trend. A spike at the second harmonic on \(f_2\) indicates a bi-annual component, seen in Fig. 6 as small asymmetries during summer and winter transitions. If intra-weekly and intra-monthly seasonality were present, we would see spikes in the periodogram around the 7th and the 12th harmonic, receptively. There are no signs thereof and we conclude that the seasonal components can sufficiently be captured by estimating the sine-cosine pairs for the first \(k=2\) harmonics⁹, effectively reducing the degrees of freedom used from \(T=365\) (excluding leap days) to \(2k + 1=5\) with less risk of losing non-seasonal information.

A two-way causal relationship between the explained and the explanatory variable can be a source of endogeneity bias when estimating the price elasticity of demand. In this context, the theoretical simultaneous determination of supply and demand is particularly vulnerable to misspecification. We evaluate the endogeneity of our initial specification with the Hausman test in Appendix B. Here, we observe that the fitted values of demand is a significant predictor for price, indicating that there is a two-way relationship between the variables.

We propose weather as fundamental source of endogeneity, a dominating determinant of natural gas demand and information that is available to all market actors. A shock in weather conditions would therefore shift both demand and the wholesale price simultaneously, imposing a classic endogeneity bias.

Examining the autocorrelation function (ACF) for daily observations of price and weather, we find that the wholesale price has a long memory, while seasonally adjusted HDD has no autocorrelation from lag 10. We therefore use the price with a ten-day lag as a proxy of the contemporaneous price, since there is no information left in HDD but a high degree of information in the lagged price.¹⁰ A contemporaneous shock to HDD would now only affect demand and not price, as information on the shock is not contained in the lagged price variable. We can therefore gain inference from using the lagged price as proxy, but avoid correlated innovations in price and demand originating from changing weather.

We employ an autoregressive distributed lag (ARDL) model, including u lags of the explained variable and \(v_i\) lags of the explanatory variable \(x_i\).The ARDL has the flexibility to handle both I(0) and I(1) variables in a single regression framework, in contrast to the procedures of Engle and Granger (1987) and Johansen (1995), restricting all variables to have the same order of integration. We can assume that we have at least one non-stationary variable, as market prices are known to contain unit roots. In addition, unit root tests are known to suffer from low power when the data-generating process is close to I(1), which may lead to misspecification.

Through the ARDL bounds test for cointegration introduced by Pesaran et al. (2001), we can test for cointegration without knowledge on the exact order of integration. In the case where a constant is included in the cointegrating relationship, the F-test is conducted on the following alternative representation of Eq. 6:The null hypothesis is thenThe test statistic from the bounds test is compared against a set of critical value bounds for the two extreme cases where all variables are I(0), the lower bound, and where all are I(1), the upper bound. If the calculated F-statistic is above the upper bound, the null hypothesis of no cointegrating relationship between the variables is rejected, whereas if the F-statistic is within the critical value band, the test is inconclusive.

If a cointegrating relationship is present, the ARDL framework enables us to separate the cointegrating level relationship from the short-run dynamics. Equation 7 can then be further reparameterized to a restricted error correction model (RECM):Here, \(\phi \) is defined as the speed of adjustment towards the equilibrium relationship in the error correction term (ECT). For convergence, \(\phi \) must be negative, significant and less than unity in amplitude. The ECT given bycan be determined from the parameters in Eq. 7 with the transformationsFrom Eqs. 8 and 9, it is clear that in equilibrium, where all the difference terms are zero, the RECM simplifies to the initial level relationship defined in Eq. 1.

Examining the results for wholesale prices, we find that demand from small consumers is highly inelastic, with a price elasticity of -0.012 both on the daily frequency and monthly frequency Table 3. This means that a 1% increase in wholesale prices would lead to a demand reduction of around 0.012%. While this impact might sound negligible, the average wholesale price rose by a remarkable 433% in the crisis year 2022 compared to a reference period of the previous sample years 2018-2021. Combined, these estimates would imply a demand reduction of \(5.3\%\), driven by wholesale prices.

For the monthly retail prices, we find an elasticity of demand of -0.037. This implies that the reaction to a relative change in retail prices is around three times as strong as the reaction to a relative change in wholesale prices. Nonetheless, the small magnitude adds to the evidence that demand is inelastic in this period, also in response to prices directly affecting the consumer.¹²

In quantities, our results imply that approximately 15TWh of the observed demand reduction in 2022 is related to the increases in retail prices. This calculation is based on the monthly price elasticity estimate¹³, which reveals by how many GWh the average daily demand changes from a 1 EUR/MWh change in the average retail price. Multiplying the coefficient with 365, we obtain the impact of a mean shift over a year. Figure 8 illustrates the impact of the observed mean shift in each variable between 2022 and the period 2018-2021.

Full size image

The largest reducing effect on natural gas demand in 2022 came from the lower number of cold days (HDD) which contributed with 16 GWh (35%) to the total reduction of 45 TWh. The higher prices also contributed a large share to the demand reduction with 15 TWh (33%). Together with the crisis awareness, the joint effect of crisis (higher prices and higher awareness) was 22 TWh which was even larger than the joint effect of weather (less HDD + higher radiation + more wind = 20.3 TWh). Clearly, this was primarily due to the large magnitude of change in the price and crisis awareness variables.

In addition, there is little uncertainty to the crisis awareness estimate and we consider it separately from the price variable. The low correlation between the crisis awareness variable and the price series (below 0.25 on both frequencies) supports the assumption that the effects of prices and general crisis awareness can be separated. An estimated demand reduction of 7TWh in 2022 from this awareness should serve as a motivation for policy makers to keep the general public informed.

While the crisis conditions prompted small consumers to decrease their consumption, our results reveal that the largest part of the demand variations is explained by weather conditions. We find that the reduction in HDD contributed to about 16TWh demand reduction in 2022 compared to the reference period. Interestingly, 2022 had the mildest winter since 2015, in contrast to the previous year which was the coldest in the same period (Eurostat, 2023c). The solar radiation from the sky was also abnormally intense, contributing to an additional heating effect on the building envelopes. In future winters, Germany might not have this double advantage, and the sensitivity to weather of small consumers must not be neglected.

Adding to the insufficient research on the pandemic effects on natural gas demand, we report no significant dummy impact from the lockdown measures.¹⁴ Only one study on price elasticity of demand, (Erias & Iglesias, 2022) includes the pandemic in their sample period, and the recent paper by Ruhnau et al. (2023) deliberately excludes the pandemic years from their sample. IEA (2020) highlights that 2020, the year of the first lockdown, was a warm year and that strong wind power generation contributed to low natural gas prices. They report a stronger decline in consumption during the lockdown but largely attributed it to the drop in demand from the industry. In line with our results, Honoré (2020) proposes that the effects of the pandemic on the consumption of natural gas were minor.

When comparing the reported price elasticity for retail prices to the literature, we observe that our result are below the majority of previous studies. Our sample period from 2018 to 2023 is characterized by extreme price volatility, during which few other studies have examined the price elasticity of demand. As an exception, Ruhnau et al. (2023) recently examines natural gas demand in a similar sample period, and provide a rough estimate on the price elasticity of demand of -0.16 given that behavioral changes in 2022 were driven by price. Similar to our findings, their calculation also indicates that the retail price elasticity of demand has been lower in recent periods.

For the price elasticity of demand on wholesale prices, the available literature is limited. Erias and Iglesias (2022), the only study to our knowledge estimating the wholesale price elasticity of demand in Europe, finds a short-run elasticity in the range of -2.2 to 0 and a long-run elasticity in the range -4.3 and -0.94 in the period 2016 to 2020. Burns (2021) estimates an average elasticity of demand of -0.09 on US data from 1980 to 2016. Although this response is stronger than our results, her estimations on subsamples in the period provide elasticity estimates close to zero. Her results support our finding that natural gas demand is highly inelastic to wholesale prices.

Obtaining lower estimates than prior literature can be attributed to the fact that our study controls for a combination of factors known to create upward bias to the price estimates. Firstly, we make deliberate modeling choices to create an appropriate control for weather effects. As the most important predictor of natural gas demand, misspecifications of the relationship between weather and demand would impact the estimate of price elasticity of demand. Secondly, we recognize the issue of spurious regression. Many prior studies have overlooked this concern, which can lead to false conclusions on the underlying relationship if the data are non-stationary. Lastly, we decompose the behavioral effect that has often been attributed to price in its entirety. Ruhnau et al. (2023) stress this issue and suggest that their own rough estimate of price elasticity of demand is inflated if an ”increase in public attention and ethical considerations” are drivers of demand reduction. Our results suggest that public awareness does have a significant impact on demand. By addressing these issues, we mitigate the risk of inflated estimates on the price elasticity of demand.

For policy makers, the wholesale price elasticity of demand can indicate whether changes in market prices cause self-regulation among small consumers. We find evidence of a reaction to price signals in the wholesale market, although it is limited in magnitude and dominated by weather effects. This finding suggests that purely price-driven policy measures are ineffective in the short run. To exploit the saving potential from small consumers during periods of restricted supply, the price hikes must either be very large as in 2022, or additional measures must be part of the policy mix. In particular raising awareness of crisis and potential scarcity can be a helpful complement to pricing signals. We show that crisis awareness alone was responsible for about 15% of the demand response in 2022.

Our results show a three times stronger response to the retail prices than to wholesale prices. This implies that the demand response from small consumers to market changes could be far greater if price signals were rapidly passed on to retail prices. However, today’s structure with yearly metering and fixed-price contracts makes such efforts impossible. In the absence of modern metering devices, there is no way of providing a fair dynamic pricing scheme without frequent information on consumption. Automatic (smart) meter reading has been widely discussed for electricity, but should also be considered for natural gas to meet the potential for increased market efficiency revealed in our results.

The consumer benefit from dynamic pricing schemes depends on the demand flexibility. Consumers are price sensitive and thus if they would see more variation in prices their demand could react and potential efficiency gains from dynamic pricing could be exploited. From an operator’s perspective dynamic pricing schemes can potentially provide for efficient gas grid management. German policy makers, during 2022, rather pointed to simple measures such as turning down the thermostat by 1 °C and drawing the curtains to save energy (Tagesspiegel, 2022). In this sense, small consumers have some short-term flexibility to adjust their heating consumption. However, a basic consumption level is needed to maintain the standard of living. If consumption is close to this basic needs level, price fluctuations directly impact either the indoor comfort of the consumers or the budget for other goods.

The welfare effect of more variable prices may vary across consumer groups. Analyzing the impact of energy price hikes in Norway, Dalen and Halvorsen (2022) find that the potential for energy saving is the lowest for low-income groups.¹⁵ As an example, they highlight that turning down the heat in unused rooms is only feasible if you have the luxury of unused rooms. This illustrates that low-income groups have a modest consumption above the levels needed to fulfill basic needs and are strongly affected by price fluctuations. This distributional effect should not be neglected when implementing policy.

Beyond the crisis situation, the price elasticity to a fossil as natural gas might provide valuable information for policy makers that intend to introduce or increase emissions pricing. Indeed, since 2021, German household consumers have to pay a CO2 tax on their natural gas consumption. The tax level increases every year in order to provide incentives to reduce natural gas consumption and do a fuel switch to renewable energies. There is no direct interference from our price elasticity estimate(s) to the effect of a \(\mathrm {CO_2}\) tax due to diverse behavioural economic issues such as tax salience etc. However, our estimates provide an indication that it might require a large emission tax for a sizeable reduction effect on demand.