Modelling, forecasting and trading of commodity spreads

摘要: Historically, econometric models have been developed to model financial instruments and markets however the vast majority of these ‘traditional’ one thing in common, linearity. While this is convenient sometimes intuitive many linear fail fully capture dynamic complex nature markets. More recently, ‘sophisticated’ methodologies evolved accurately ‘non-linear’ relationships that exist between time series. This rapidly advancing field quantitative finance known as Artifical Intelligence. The earliest forms artificial intelligence are Neural Networks since using more accurate learning algoirthms. networks also particular use because their capability being able continually learn new information fed into network. In research data introduced both fixed sliding window approaches for training each networks. Futhermore, Genetic Programming Algorithms highly regarded industry increasingly applied an optimisation technique. Therefore, non-linear supported by existing a result become practical tools optimising predicting future movements assets. absence computational algorithms rationalise large amounts data, investors confronted with difficult seemingly impossible task trying comprehend datasets information. Nevertheless, advancements computing technology enabled market participants benefit from neural (NN) genetic programming (GP) order optimise identify patterns trends explanatory variables target outputs. importance agricultural such grains, precious metals other commodities informationally rich readily available evaluate. Among first analysis were Rumelhart McClelland (1986), Lippman (1987), Medsker et al. (1993). extensively foreign exchange (Hornik al., 1989; Lawrenz Westerhoff, 2003), credit (Tam Kiang, 1992), volatility forecasting (Ormoneit Neuneier, 1996; Donaldson Kamstra, 1997), option pricing, (Hutchinson al.,1994), portfolio (Chang 2000; Lin 2001), (Swales Yoon, 1992) emerging (Kimoto 1990) stock markets, technical trading rules (Tsai al.,1999; Neely 2003). application futures contracts inparticular, commodity spread trading, limited. Trippi DeSieno (1992) Kaastra Boyd (1995), among explore apply forecast Financial assets influenced array factors including but not limited to; human behaviour, economic variables, systematic non-systematic . As result, academics practioners devised numerous explain series fundamental analysis, behavioural finance. purpose identify, trade daily changes spreads combination novel nonlinear modeling techniques performance enhancing filters. During process, used expansive datasets. Each produce predictions periods. forecasts t+1 horizons examined. Progressively, chapter presents evolution area address inefficiencies associated traditional architectures. total collection five proposed analysed ‘spreads’. These benchmarked against which include Naive strategies, Moving Average Convergence Divergence (MACD) buy hold Autoregressive (ARMA) models, Cointegration models. final mixed approach employed outputs benchmark inputs during includes various adaptations programming. Through historical input, methodology trained construct ‘optimal’ Models selected Exchange Traded Funds (ETFs) Futures contracts. all cases reader presented results unfiltered filtered simulations. aim thesis hedgers speculators who interested applying spreads. By allowing input can valuable tool Empirical evidence reveals statistically superior compared they higher risk adjusted returns. Moreover, output dataset train ‘synergy’ ‘mixed model’ approach. improve offers examples filters be speculators. On whole contributes wealth knowledge academic studies it conclusive support widespread integration modelling form intelligence. evaluated statistical measures well widely institutions.