Miðvikudaginn 28. maí kl. 14:00 mun Þorvarður Snær Örnólfsson verja 30 ECTS verkefni sitt til meistaragráðu í fjármálaverkfræði „Information and Predictability in Financial Time Series"

• Stofa: M102 og eru öll velkomin.
• Nemandi: Þorvarður Snær Örnólfsson
• Leiðbeinandi: Sverrir Ólafsson
• Prófdómari: Jón Guðnason

Útdráttur: This thesis explores whether information–theoretic measures, in particular mutual information (MI), can reveal predictive structure in financial markets, not captured by classic linear statistics. Daily data for four assets, the S&P500, the OMX Iceland index, the BTC/USD exchange rate, and the USD/ISK exchange rate, covering the period 2006–2025 are analysed. Using the Kraskov k-nearest‑neighbour estimator, lag‑wise MI is computed up to 30 trading days, while phase‑randomised surrogates provide 95% confidence bands that preserve all linear autocorrelations and destroy nonlinear dependence. The S&P500 exhibits significant nonlinear memory out to twelve lags, whereas the other asset MI curves fluctuate around estimator noise. Fitting an exponential decay model yields half‑lives of approximately 9.6 days for the S&P500, 5.2 days for USD/ISK, and below one day for OMXI and BTC/USD. Rolling estimates show that half‑life lengthens during crisis periods, mirroring volatility clustering. Lagged technical indicators contain measurable information for next day S&P500 returns but offer little benefit for the other assets. These results demonstrate that information‑theoretic tools complement linear methodologies.