The proposal deals with feedback control based methods for investment in lieu of the traditional financial models. The goal is to exploit the properties of feedback which do not require elaborate price modeling. Consistent with basic feedback principles, the controllers are reactive to price rather than predictive in nature.
Intellectual Merit: The research will stimulate the application of feedback principles in finance as an alternative to economic modeling. Rather than using unreliable stochastic price models, we treat price as an external uncontrolled input against which we seek robust performance. The feedback controller simply processes the history of price and possibly volume to determine the appropriate level of investment. In a portfolio context, controller design will result in new alternatives to the celebrated Markowitz theory for weighting of various components. The proposed research includes the notion of robust certification of performance in a so-called idealized market. The approach is to replace expensive and time consuming back-tests with a new robustness paradigm for evaluating investment strategies.
Broader Impact: The research plans to develop simple recipes and software for the self-directed investor who is in essence competing against well capitalized hedge funds and large financial institutions that can afford to take risky bets with potentially high payoffs. We envision workshops for non-experts aimed to teach attendees how to manage investments, a web site from which tutorial literature and trading codes can be readily downloaded. This line of research has the potential to motivate a new generation of graduate students in systems engineering to consider careers in finance.
An important purpose of this exploratory research under the EAGER program was to bring both new and classical tools from the area of feedback control systems into the area of finance. In this regard, problems involving investment were studied via our new paradigm exploiting adaptive feedback loops in the context of investment decision-making. A very simple special case of the type of investment set-up which we studied is depicted in the image accompanying this report. By recasting investment problems into the context of a feedback control loop, new tools can be brought to bear. One major outcome of the research was the development of new approaches to investment which require little or no mathematical modelling of the evolution of asset prices; no predictions about the future are used. Instead, our research concentrated on taking advantage of the "adaptive" properties of a closed-loop feedback system. To provide a crude analogy which helps understand the type of model-free investment schemes which we developed, one might imagine the problem of bringing bathtub water to a desired temperature set point. Faced with such a problem, many would ridicule the notion that expertise is required in thermodynamic fluid modelling. With an adequate sensor, namely a hand in the water, and adequate actuator, namely a hand on the faucet, the following simple adaptive feedback rule will generally suffice: "If the bathtub water is too hot, add cold water. If the bathtub water is too cold, add hot water.'' An adaptive feedback loop based on this simple temperature trend-following principle will typically be nimble enough to handle rather unpredictable water inflows. In all of our research publications, presentations and other publicly disseminated work involving financial investment, ideas analogous to bathtub filling were in play. Whereas adaptive control for the bathtub is achieved via adjustment of inflows and outflows based on the water temperature trend, in the case of financial markets, the amount of money invested in a stock is adapted over time based on the profit-loss trend. Our research was both widely disseminated and broadly publicized. One highlight, available to the public, is the so-called Bode Lecture presented by the Principal Investigator to a 1000+ audience of researchers at the 2013 IEEE Conference on Decision and Control; see www.ieeecss-oll.org/lecture/can-control-science-bring-new-insights-stock-trading-research-0 Given the multi-disciplinary aspect of this research project, we actively presented and disseminated our results to both the finance and control systems areas. This was accomplished via a large number of invited lectures, conference presentations and publications and tutorial sessions at the American Control Conference and the IEEE Conference on Decision and Control. Finally, to describe the largest potential impact of the work going forward, it is important to mention one of the main sources of motivation for this research. Namely, the pursuit of these new methods is motivated by the significant failure of many classical financial models over the decade covering 2000-2010. Dramatic and sudden changes in market volatility and asset correlations rendered classical models in finance rather useless and in the market crash of 2008, ordinary investors who trusted asset managers were victims of failures of modern portfolio theory. Given these facts, one of the major outcomes of the EAGER research is the possibility that new and more robust methods of portfolio management will be developed which are minimally reliant on price modeling.
