Investments unwrapped : demystifying and automating technical analysis and hedge-fund strategies

In this thesis we use nonlinear and linear estimation techniques to model two common investment strategies: hedge funds and technical analysis. Our models provide transparent and low-cost alternatives to these two nontransparent, and in some cases prohibitively costly, financial approaches. In the case of hedge funds, we estimate linear factor models to create passive replicating portfolios of common exchange-traded instruments, that provide similar risk exposures as hedge funds, but at lower cost and with greater transparency. While the performance of linear clones is generally inferior to their hedge-fund counterparts, in some cases the clones perform well enough to warrant serious consideration as low-cost passive alternatives to hedge funds. In the case of technical analysis - also known as "charting" - we develop an algorithm based on neural networks that formalizes and automates the highly subjective technical practice of detecting, with the naked eye, certain geometric patterns that appear on price charts and that are believed to have predictive value. We then evaluate the predictive ability of these technical patterns by applying our algorithm to stocks and exchange rates data for a number of stocks and currencies over many time periods, and comparing the unconditional distribution of returns to the return distribution conditional on the occurrence of technical patterns.