Level E

		client area

Technology
MAYA™ uses thousands of predictive features to model financial time-series data. The most robust and historically informative features are automatically selected, and used as inputs to Level E's proprietary financial market models. The financial features are chosen to be maximally independent and robust across multiple time scales and fluctuations in the data. Features that are maximally informative about future changes in the market are the most useful features, so after a principled and thorough characterization of the robustness of these features, the most useful, and reliable can be selected automatically. The market models are adaptive in real-time, to allow for changes in the market environment as well as being adaptable to other markets.

An array of Machine Learning (ML) methods are used to predict the performance of various stocks, producing a set of return predictions for multiple model-stock pairs, and recommended investment actions. The stochastic nature of the market is addressed formally using ML methods which produce not only a predicted change in stock value, but also a formal confidence level of their prediction. The models are trained on recent financial history, and are continuously updated as more data becomes available to keep the models relevant to the current market climate. Furthermore, changes in the market regime can be detected by analyzing the investment trends and market distributions, allowing more relevant models to be activated, or current models to be updated appropriately.

An investment portfolio is constructed using the signals produced by the array of predictive models. Model predictions are combined with the need to ensure diversification of exposure, and taking into account additional client constraints. Further factors can be incorporated at this stage by fine tuning the execution strategy. For example the effect that the time of day has on the distribution of stock prices can be exploited to optimize the investment strategy.

The models developed by Level E represent the most recent and promising developments in the fields of Machine Learning, Data Mining and Probabilistic AI. These formal mathematical approaches to modeling stochastic data are particularly suited to the intrinsic uncertainty of financial markets, and are specifically designed to capture structure within noisy data. There have been many recent practical advances in approximate inference and efficient learning methods for such models, which can be used for real-time applications where the size of the model and the complexity of the data is prohibitive for naive exact inference.

As computers become faster and cheaper, formal probabilistic models become more practically applicable. Within the past 10 years many previously intractable models have been used in various fields of research, from speech recognition to climate modeling. Financial data is ideal for probabilistic modeling, as it requires high noise tolerance and inference of complex underlying models that couple various stock prices together. In addition to formal probabilistic modeling, more traditional AI approaches are also employed where the system is well characterized, or when the complete inference of the equivalent probabilistic model would be prohibitive.

Home

Disclaimer