Assessing the Market Efficiency of Carbon Prices in the European Union Emissions Trading System: A Forecasting Approach

Abstract

Denne avhandlingen implementerer både univariate og multivariate tidsrekke- og maskinlæringsmodeller for å vurdere markedseffektiviteten til karbonpriser innenfor EU ETS. Modellene inkluderer en ARMA-modell, en multippel lineær regresjon for tidsrekker, samt en univariat og multivariat LSTM-modell for å forutsi avkastningen på karbonpriser. Resultatene våre avslører at prognoseytelsen betraktelig forbedres ved å inkludere flere finansielle variabler, hvor den multivariate LSTM-modellen gir den beste ytelsen i form av RMSE. Videre bruker vi forklarbar kunstig intelligens med SHAP-verdier for å identifisere økonomisk usikkerhet som den viktigste variabelen for prognoseytelsen. Når det gjelder markedseffektivitet, vurderer vi prognoseytelsen mot de svake og semi-sterke formene for markedseffektivitet. Våre funn antyder, gitt dataen brukt, at den svake formen for markedseffektivitet opprettholdes. Imidlertid er resultatene våre ikke tilstrekkelig for å bekrefte eller avkrefte den semi-sterke formen for markedseffektivitet, da ytteligere indkludering av variabler potensielt kunne forbedret prognoseytelsen. Basert på vår tilnærming og våre funn, konkluderer vi derfor med at karbonmarkedet innenfor EU ETS er effektivt.

This thesis implements both univariate and multivariate linear time series and machine learning models to assess the market efficiency of carbon prices within the European Union Emissions Trading System. Specifically, the models include an autoregressive moving average model, a multiple linear regression for time series model, and a univariate and multivariate long short-term memory model for forecasting carbon price returns. Our results reveal that out-of-sample forecasting performance is substantially improved by including additional financial variables, with the multivariate long short-term memory model yielding the best performance in terms of root mean squared error. Furthermore, using explainable artificial intelligence with SHapley Additive exPlanations, we identify economic policy uncertainty as the most important feature for forecasting performance. In terms of market efficiency, we evaluate the forecasting performance towards the weak and semi-strong forms of market efficiency. Our findings suggest, given the data used, that the weak form of market efficiency holds. However, our results fail to neither confirm nor deny the semi-strong form of market efficiency, as further inclusion of additional data could potentially improve forecasting performance. Thus, given our approach and findings, we conclude the carbon market within the European Union Emissions Trading System as efficient.