Electrification & AI Demand Innovations and Real LME Forward Returns
Authors
Borrild, Frederik Hylander ; Houe, Mads Plougmand ; Kaad, Oliver
Term
4. semester
Publication year
2026
Pages
85
Abstract
This thesis tests whether equity-market and real-economy proxies for demand related to electrification and artificial intelligence contain information about real three-month forward returns on the London Metal Exchange for six industrial metals. Two econometric models address this: Model A estimates same-period associations by regressing real forward returns on orthogonalised thematic innovations, and Model B assesses one-month-ahead predictability conditional on a Fama-style forward-basis benchmark. The clean sample comprises 179 monthly observations from September 2010 to July 2025. In-sample estimation uses OLS with Newey-West standard errors and general-to-specific selection, while out-of-sample performance is evaluated via recursive forecasting and Clark-West tests. Contemporaneous results support an electrification channel: coefficients are positive across all six metals, statistically significant at the 5 percent level for aluminium and zinc and at the 10 percent level for copper, nickel, and lead, with tin positive but imprecisely estimated. The AI proxy is not significant and shows no consistent cross-metal pattern. Sub-samples indicate the electrification link is stronger earlier in the sample and attenuates after 2016, consistent with time-varying information content. Inventory mediation is small in the contemporaneous specification, pointing to an expectations rather than immediate physical-tightness channel. Predictive evidence is weak: ETF-based proxies do not improve one-month-ahead forecasts beyond the basis benchmark; out-of-sample R-squared values are negative and Clark-West tests do not reject the benchmark. A parallel extension replacing the ETF proxies with orthogonalised EV-sales innovations yields stronger signals for selected metals, with rejections for nickel and tin, but this remains a robustness check and does not adjust for data-release lags. Overall, forward markets appear to incorporate observable thematic demand information within the month, leaving limited reliable monthly forecastability.
Dette speciale undersøger, om aktiemarkeds- og realøkonomiske proxyer for efterspørgsel knyttet til elektrificering og kunstig intelligens indeholder information om reale tre-måneders forward-afkast på London Metal Exchange for seks industrimetaller. To økonometriske modeller adresserer spørgsmålet: Model A estimerer samtidige sammenhænge ved at regressere reale forward-afkast på ortogonaliserede tematiske innovationer, og Model B vurderer én-måned-fremme forudsigelighed betinget af en Fama-inspireret forward-basis-benchmark. Datasættet omfatter 179 månedlige observationer fra september 2010 til juli 2025. In-sample estimering anvender OLS med Newey-West standardfejl og general-to-specific modelvalg, mens out-of-sample præstation vurderes via rekursive prognoser og Clark-West-tests. De samtidige resultater støtter en elektrificeringskanal: koefficienterne er positive for alle seks metaller, signifikante på 5 pct.-niveauet for aluminium og zink og på 10 pct.-niveauet for kobber, nikkel og bly, mens tin er positiv men upræcist estimeret. AI-proxyen er ikke signifikant og udviser intet konsistent mønster på tværs af metaller. Delprøver indikerer, at elektrificeringssammenhængen er stærkere tidligt i perioden og aftager efter 2016, i tråd med tidsvarierende informationsindhold. Lagerformidling er lille i den samtidige specifikation, hvilket peger på en forventningskanal frem for akut fysisk stramhed. Den forudsigende evidens er svag: ETF-baserede proxyer forbedrer ikke én-måned-fremme prognoser ud over basis-benchmarket; out-of-sample R2 er negative, og Clark-West-tests afviser ikke benchmarket. En parallel udvidelse, hvor ETF-proxyer erstattes af ortogonaliserede elbilsalgsinnovationer, giver stærkere signaler for udvalgte metaller, med afvisninger for nikkel og tin, men dette behandles som en robusthedstest og tager ikke højde for datoforsinkelser. Samlet tyder resultaterne på, at forwardmarkeder indpriser observerbare tematiske efterspørgselsinformationer inden for måneden, hvilket begrænser pålidelig månedlig forudsigelighed.
[This apstract has been generated with the help of AI directly from the project full text]