The time horizon of decision-making is an essential dimension of economic problems but is difficult to explicitly define. In this thesis, we use time series analysis augmented by wavelet transform methods to precisely identify distinct time horizons in economic data and measure their explanatory power. This enables us to address three timely and persistent questions in the literature on commodity derivatives markets are addressed. First, are findings of long memory (fractional integration) in commodity futures price volatility spurious, following Granger?s conjecture? Yes, only two out of eleven commodities are characterized by true long memory and certain stochastic break models (e.g. Markov-switching) are found to be more plausible. Second, do large Index Traders such as commodity pools and pension funds increase futures price volatility through a large volume of trading activity? This appears to be true only for non-storable commodity contracts. Third, can we improve the accuracy of term structure models of futures prices by (i) including more state variables to better capture maturity and inventory effects, and (ii) filtering out what appears to be noise at the shortest time horizons? The results suggest that (i) three state variables is an optimal choice and (ii) estimates using filtered data are not improved and the noise may be economically meaningful.