Forecasting the Nordic Energy Sector : An Evaluation of the Efficient Market Hypothesis

This thesis investigates the weak form of the efficient market hypothesis (EMH), by examining the predictive power in historical data in short-term forecasting. An ARIMA model is constructed to represent classical statistical modeling in the matter, whilst a LSTM model is constructed to represent more complex machine learning modeling. Both models are tested against a naïve random walk by forecasting closing prices on the Nordic energy sector. The results show that the ARIMA model achieves predictive accuracy which is statistically indistinguishable from the random walk, indicating that the forecasting errors are not positively impacted by modeling linear autocorrelation structures. The LSTM model performs worse overall than both the ARIMA and the baseline model, although two cases where it outperforms its counterparts are acknowledged. The findings suggest that inconsistent and unreliable patterns which can be exploited short-term exists in financial markets. Overall, the study nonetheless shows results aligning with the efficient market hypothesis in its weak form, showing that historical values holds little predictive power in short-term financial forecasting. Thus, complex forecasting models solely reliant on historical values add few practical benefits in this regard. Future research should explore broader hyperparameter tuning and use of longer prediction horizons, mainly to take more advantage of the LSTMs potential capacity.