Algorithmic Trading on German Electricity Markets

Project Overview

This project leverages ensemble tree based machine learning models to predict the direction of the spread (imbalance vs. day-ahead prices) in the German Electricity Market, with the goal of maximizing cumulative Profit & Loss (P&L).

The dataset consists of 15-minute interval observations with the primary features being:

• Wind generation forecasts - Impact supply variability from renewable sources.
• Solar generation forecasts - Capture solar energy’s contribution to the electricity grid.
• Load (demand) forecasts - Reflect consumer demand patterns that influence electricity prices.

Motivation

The spread reflects the dynamics of electricity supply and demand:

• When supply exceeds demand, imbalance prices tend to decrease to account for the excess.
• When demand surpasses supply, imbalance prices generally increase.

By modeling these relationships, the project aims to develop a systematic trading strategy to exploit these price differences.

Instructions

Clone the repository:

git clone https://github.com/anthonymakarewicz/algo-electricity-trading.git electricity_trading
cd electricity_trading

Usage

1. Exploratory Data Analysis (EDA)

Notebook: eda.ipynb

• Performs an in-depth analysis of the dataset:
  • Analysis over several frequencies (Daily, Weekly, Monthly)
  • Outlier Detection using Boxplots
  • Seasonality analysis using line plot and Autocorrelation Function

---

2. Data Processing

Notebook: data_processing.ipynb

• Handles outliers using winsorization and hampel filter.
• Compares and applies several transformations techniques:
  • Log Transformation
  • Box-Cox Transformation
  • Yeo-Johnson Transformation

---

3. Feature Engineering & Selection

Notebook: feature_engineering_selection.ipynb

• Generate new features to extract relevent information including:

  • Time/Dates features
  • Cyclical features
  • Interaction effects
  • Rolling features

• Leverage various feature selection techniques to remove irrelevent features including:

  • Correlation Filter
  • Recursive Feature Elimination (RFE)
  • Embedded Methods (Random Forest Feature Importance)

---

4. Model Evaluation

Notebook: model_evaluation.ipynb

• Trains and evaluates ensemble models for spread prediction such as:
  • XGBoost
  • LightGBM
  • Random Forest
• Perform hyperparameter tuning using Grid Search to avoid overfitting
• Compares models using cumulative profit & loss (P&L) curves.