Abstract: In this paper, a multivariate constrained robust M-regression (MCRM) method is developed to estimate shaping coefficients for electricity forward prices. An important benefit of the new method is that model arbitrage can be ruled out at an elementary level, as all shaping coefficients are treated simultaneously. Moreover, the new method is robust to outliers, such that the provided results are stable and not sensitive to isolated sparks or dips in the market. An efficient algorithm is presented to estimate all shaping coefficients at a low computational cost. To illustrate its good performance, the method is applied to German electricity prices.
Introduction
In the volatile world of electricity markets, "shaping" is a critical task. It involves taking a low-granularity price—like a contract for power delivery over an entire year—and breaking it down into higher-granularity components, such as quarterly, monthly, or even hourly prices. This paper introduces a sophisticated new tool for this task: Multivariate Constrained Robust M-Regression (MCRM).
The Challenge of Granularity and Arbitrage
Electricity prices are notoriously seasonal and prone to "sparks" (sudden, extreme price spikes) or "dips." Traders and risk managers need to "shape" forward curves to value complex derivatives or manage portfolios. However, traditional shaping methods face two major hurdles:
- Model Arbitrage: If quarterly prices are estimated independently, their average might not equal the yearly price they were derived from. This creates arbitrage opportunities within the model, a significant flaw for financial applications.
- Sensitivity to Outliers: Standard regression techniques like Least Squares attempt to fit every data point. In electricity markets, a single extreme weather event or market glitch can pull the entire curve out of alignment, leading to unstable valuations.
As shown in the table above, the market provides prices for a wide variety of overlapping delivery periods, from daily (D) and weekly (W) to quarterly (Q) and yearly (Y) contracts. Reconciling these different scales while maintaining internal consistency is the core difficulty of the shaping problem.
The MCRM Methodology
The authors propose a Multivariate Constrained Robust M-Regression (MCRM) framework to address these issues simultaneously. The method relies on three pillars:
- Multivariate Framework: Instead of fitting one time period at a time, the model treats all subdivisions (e.g., all four quarters of a year) as a single vector, allowing for simultaneous estimation.
- Linear Constraints: The model includes a penalty term in its cost function that forces the estimated coefficients to obey a "no-arbitrage" rule. This ensures the sum of the parts always equals the whole.
- Robust M-Estimation: Instead of minimizing squared errors, the method uses a bounded loss function (specifically the Hampel function). This effectively "downweights" observations that deviate significantly from the main market trend, ignoring isolated sparks.
The solution is implemented via an efficient Iteratively Reweighted Least Squares (IRLS) algorithm, making it fast enough for real-time trading applications.
Empirical Results and Stability
The methodology was validated using real-world data from the German Baseload Power market (EEX exchange), using daily quotation prices from 2004 to 2015 for training and the start of 2016 for out-of-sample testing.
Key findings include:
- Superior Stability: The MCRM method provided a better fit for the majority of the data by ignoring noise. The model naturally flagged periods like the 2009 financial crisis as "atypical" without manual intervention.
- Out-of-Sample Performance: When tested on 2016 data, the robust MCRM method significantly outperformed industry-standard approaches, such as the Fleten and Lemming (2003) model.
- Arbitrage Elimination: The penalty parameter proved highly effective; the model reached a stable, arbitrage-free state rapidly as the penalty increased.
Conclusion
By solving for all coefficients at once under a constraint, the no-arbitrage condition becomes a fundamental part of the model's geometry rather than a post-processing step. While the current model uses "soft" constraints and univariate regressors, it provides a robust foundation for future work, including the incorporation of external factors like coal, gas, or carbon prices directly into the shaping process.
Summary updated