Dynamic Adaptive Portfolio Allocation Theory
Dynamic Adaptive Portfolio Allocation Theory (DAPAT) is a theory and set of practices that i would like to introduce. It is a term that describes the proprietary method used to develop investment strategies at CSS Analytics. DAPAT incorporates the best concepts from complementary theories and disciplines and integrates them in a meaningful way:
The pillars of modern portfolio theory rest on mean-variance optimization concepts introduced by Markowitz. The central usefulness of this theory is in the “free lunch” introduced by diversification amongst assets that are not perfectly correlated. The goal is to optimize the arithmetic mean relative to the standard deviation.
Gambling theory as employed by card counters, rests on the Kelly Criterion. This is a statistical measure which maximizes the geometric mean while simultaneously being constrained by the probability of going bankrupt being estimated to be zero. It is based on win/loss ratios and the estimated percentage of winning hands/ or trades in the case of investing.
Money management theory as popularized by Van tharp, emphasizes the normalization of position sizing based on volatility and the use of stop losses to reduce risk. The importance of these concepts applied to trading, is to show the ability of randomly generated strategies to make money comparable, or perhaps even with better risk-adjusted results than buy and hold investing.
Machine learning, neural networks, and non-parametric statistical methods, are designed to learn and adapt from historical data. This can be accomplished even if the data is chaotic or non-normal. This allows investment strategies to be created entirely by data mining, and also allows existing strategies to survive rapidly changing environments. It also permits the processing of tremendous amounts of data efficiently and quickly that would be impossible for other methods to accomplish.
Robust statistics is a separate discipline that emphasizes the use of robust methods as a powerful tool to increase the reliability and accuracy of statistical modelling and data analysis. It helps us understand the limits of backtesting as well as optimization, and prescribes solutions that permit the effective usage of these tools.
All five disciplines can be combined by merging their different points of strength with their different points of weakneses. A portfolio that employs the principles underlying all of these disciplines will survive in the new trading landscape. Those that do not, will produce unpredictable and diminishing results over time.
I will build substantially on this theory throughout the course of my blogs, hopefully i can shed some light on how combining these disciplines will lead to far superior results.