Modelling the Performance of Investment Strategies. Concepts, Tools and Examples

摘要

We present a predictive model of investment strategy performance applied to individual assets or protfolios of assets. We introduce the concept of "meta modelling' where the modelled quantity is investment strategy performance rather than market returns. An an example, we propse to model and and forecast the performance of trend following strategies. When applied to relative stock prices the return (i.e. profit/loss) of such strategy has a straightofrward interpretation as a measure of the stock's relative cyclicality. We develop a critieria which links predictive power to incremental return. Two models will be used on randked and unranked data and their results will be be analysed and compared. The first model is a linear regression, the second model is a multilayer perceptron trained with a standard back-proopagation algorithm. Both models provide small but consistent incremental value with the neural network outperforming its lienar ocunterpart by a factor of 2. This usggests presence of nonlinearities probably induced by the interactions between some of the technical, economic and financial variables and which are undetectable b standard econometric analysis. The dataset is compsoed by financial data on 90 French Stocks drawn from the SBF250 index, and by economic indicators. Both models will be used on several protions of the dataset in order to estimate the generalisation capability of the models across time and stocks. This study is also an opportunity to test profitablity of MA strategies appleid on relative stock prices and pressents a methodology to improve the economic performance of usch strategies.