How are Day-ahead Prices Informative for Predicting the Next Day's Consumption of Natural Gas? Evidence from France.

AuthorThomas, Arthur

    The accuracy of the gas demand forecasts issued by the Transmission System Operator (TSO) is now becoming an important matter in regulatory debates and has motivated the adoption of dedicated incentive schemes in some countries. (1) In response, TSOs have implemented advanced forecasting tools combining several methodologies (e.g., time series, neural network, adaptative logic networks) along with a plethora of variables (e.g., temperatures, wind speeds, rain, snow, cloud cover, forecasted power demand). Yet, despite these efforts, forecasting the next day's consumption of natural gas remains a challenging task. (2) By testing an alternative forecasting approach using the information in day-ahead prices, the present paper usefully contributes to the ongoing discussion on the performance of short-term consumption forecasts used in the gas industry.

    Over the last two decades, a series of European regulatory reforms have prompted the emergence of a collection of day-ahead wholesale markets for natural gas, the so-called "gas hubs," that turned out to become an important source of gas procurement as the previously monopolized industry structure gradually became more fragmented (Miriello and Polo, 2015). By construction, these markets have been developed to cope with local network balancing needs and allow an optimal scheduling of resources. Their functioning is thus closely affected by the detailed balancing rules used by the TSO. An important milestone in the design of these balancing procedures occurred in 2014 when the European Commission imposed a unified network code on TSOs. (3) Yet, despite that harmonization, market analysts recurrently point to significant differences in the perceived degree of trading liquidity observed at the European gas hubs (Heather and Petrovich, 2017). Thus, a fundamental public policy issue is whether the current market design generates transparent spot prices that reflect the market participation of all concerned economic agents (suppliers, trading firms, and consumers).

    In the electricity sector, Forbes and Zampelli (2014) proposed an original approach to examine the informational content of day-ahead electricity prices. They hypothesized that if day-ahead markets for electricity were efficient then these prices should reflect the processed information of all market participants regarding the next day's load. That consideration led them to test whether it was possible to improve the predictions of the next day's electricity load using only the information contained in the day-ahead price series. The authors examined California's PG&E aggregation area and applied traditional time-series techniques (namely a linear ARMAX specification) to model the next day's load as a function of a single explanatory variable: the day-ahead spark ratio defined as the electricity to gas price ratio. Their results reveal that this approach is sufficient for computing very accurate forecasts which outperform those published by the system operators. Remarkably, their results document the major informational content of day-ahead prices in the case of electricity.

    Our paper is the first econometric study of the daily interactions between day-ahead prices and the natural gas demand observed at a given hub. In some respects, it extends Forbes and Zampelli (2014)'s approach in highlighting what necessary specific dimensions must be considered to produce accurate natural gas demand forecasts. However, we acknowledge that dealing with gas prices requires modeling specificities that differ from the ones used for electricity. Essentially, three main characteristics are typical of the gas market: (i) the fact that the aggregate gas demand emanates from both end-users and thermoelectric generation, (ii) the expected nonlinearity in the relationship between price and demand incidental to the level of the relative price of electricity to natural gas (spark ratio), and (iii) the time series (unit-root) properties of the data. We further elaborate on these three distinctive features in section 2 as they provide the essential justifications for our modeling choice and are thus key in our analysis. In light of these features, we consider two nonlinear specifications that are extensions of the well-known Autoregressive Distributed Lag (ARDL) model: the Nonlinear ARDL (NARDL) and the Threshold ARDL (TARDL), that we propose here, and that is a straightforward extension of the genuine ARDL model. By doing so, we investigate the presence of a long-run relationship between consumption, day-ahead price, and spark ratio and explore the potential asymmetric influence of the spark ratio on observed consumption levels. (4)

    Our application deals with two French wholesale markets--namely, the Point d'Echange de Gaz Nord (PEG Nord) and the Trading Region South (TRS)--over the period 2015-2018. This allows us to present a series of original findings. First, we provide evidence of a symmetric and significant long-run relationship between the daily demand level, the spot price of natural gas, and the relative price of electricity to gas. Second, we document the magnitude of the reaction of daily gas consumption to the price of natural gas in the short run. Third, we show that, in the short run, the spark ratio has an asymmetric and nonlinear impact on observed demand levels. In each market, the reported relationship obtained with the TARDL model is sufficiently robust to producing day-ahead forecasts that are considerably more accurate than those published by network operators.

    Our main contribution to the literature is to show that publicly available information, such as day-ahead prices, can be used to produce efficient forecasts of tomorrow's consumption by relying on a quite simple econometric model. The fact that our demand forecasts are much better than those provided by TSOs appears quite puzzling as TSOs are expected to hold superior information. Why such superior information does not translate into better demand predictions remains an open question.

    Our framework can provide useful guidance to a large audience interested in the dynamics of natural gas demand in the short run and in the reaction of that demand to market prices. While a large literature in applied econometrics literature has approached the question using medium to low frequency data (e.g., monthly, quarterly or annual), that reaction has never been examined using daily data. In principle, the use of daily data is much more relevant for eliciting the short run effects from lagged changes in prices on the observed demand. Geweke (1978) stresses that estimation over broader data intervals can result in significant bias. His analysis indicates that aggregation over time can create some kind of omitted variables bias problem because the intertemporal lag distribution is not properly specified. In our case, the use of daily data may provide more reliable estimates of the marginal impacts that gas and electricity prices have on natural gas demand in both the long- and the short-run, and thus on the dynamics of the reactions of natural gas demand to energy price changes. As these marginal impacts play an important role in the models developed to examine the effects of a possible sudden temporary disruption in gas supplies on optimal import policies, (5) our modeling approach usefully contributes to the policy discussions related to the security of foreign-controlled gas supplies in importing nations.

    Though our discussion is confined to the French case, we believe that the results are pertinent for other countries engaged in a transition toward less carbon-intensive energy systems. In France, the gas consumption emanating from the power sector exhibits large and sudden variations because Combined Cycle Gas Turbines (CCGT) plants are primarily dispatched as peaking units, which leads to large flow variations in the gas network as these plants ramp up and down. That situation is likely to prefigure the new role assigned to gas-fired power plants when a previously thermoelectric dominated power system experiences a massive penetration of renewable generation. Because of their almost zero marginal costs of production, solar and wind generators are placed at the beginning of the electricity 'merit order,' which greatly reduces the need to dispatch gas-fueled generators as baseload or mid-merit units (Green and Vasilakos, 2010) and thus leads to large variations in the gas demand emanating from these plants (Qadrdan et al., 2010).

    The present analysis has important implications for the cost-efficient operation of a natural gas pipeline system and its economic regulation in the broader context of decarbonized energy systems. As gas-fired generation is increasingly used as a backup technology in power systems, the short-term variability observed in the power sector is increasingly transferred to gas infrastructures. Because of both the increased variability of the next days' loads and the possibility of demand forecasting errors, gas TSOs' are forced to adopt precautionary network management strategies based on the buildup and discharge of a pipeline inventory named linepack (Gopalakrishnan and Biegler, 2013; Tran et al., 2018). The linepack storage is an important source of short-term flexibility that could be extremely valuable in decarbonized energy systems as shown in Sun et al. (2012) and Arvesen et al. (2013). Yet, from a market design perspective, that storage service is seldom sold to market participants in Europe and a substantial share of its cost is socialized by means of transport tariffs (Keyaerts et al., 2011; Hallack and Vazquez, 2013). Because of these factors, there is a heightened interest in the accuracy of the gas demand forecasts published by the TSOs. That topic is now considered to be an important regulatory policy issue that has recently motivated the implementation of specific incentive schemes in some countries (e.g...

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT