- The Book Structure
- Strategies Considered in This Book
- Scientific and Empirical Approaches to Developing Automated Trading Strategies
- Rational Approach to Developing Automated Trading Strategies
This chapter is from the book
This chapter is from the book
This chapter is from the book
Scientific and Empirical Approaches to Developing Automated Trading Strategies
There are two main approaches to the development of automated trading strategies. The first approach is based on the principles and concepts defined by a strategy developer. All the elements composing such a strategy originate from economic knowledge, fundamental estimates, expert opinions, and so forth. Formalization of such knowledge, estimates, and assumptions in the form of algorithmic rules provides a basis for creating an automated trading strategy. Following the example of Robert Pardo, we will call this a scientific approach.
At its extreme, the scientific approach provides for a total rejection of optimization procedures. All the rules and parameters of a trading system are determined solely on the basis of knowledge and forecasts of the developer. Apparently, the likelihood of creating a profitable strategy, while avoiding engagement in optimization procedures, is extremely low. Scientific approach in its pure form is hardly applicable in real trading.
The alternative approach is based on the complete denial of any a priori established theories, models, and principles while developing automated trading strategies. This approach requires extensive use of computer technologies to search for profitable trading rules. All algorithms can be tested for this task (with no concern for any economically sound reasons standing behind their application). Candidate algorithms can be selected from a number of ready-made alternatives available or actually constructed by the system developer. The method of algorithm creation is not determined by preliminary assumptions and is not limited by any exogenous reasoning. Trading rules are selected solely on the basis of their testing using historical data. The resulting strategy is devoid of any behavioral logic or economic sense. Following Robert Pardo, we will call it an empirical approach.
At its extreme, the empirical approach is a purposeful quest for algorithms and parameters that maximize simulated profit (minimize loss or satisfy any other utility function). This approach is based exclusively on optimization. Nowadays there is a wide choice of high-technology software that facilitates fast development of effective algorithms and provides for establishment of optimal parameter sets. For example, neuron networks and genetic methods represent powerful tools that enable relatively fast finding of optimal solutions through the creation of self-learning systems.
Usually trading strategies constructed on the basis of the empirical approach show remarkable results when tested on historical time series, but demonstrate failure in real trading. The reason for this is overfitting. Even walk-forward analysis does not eliminate this threat because the significant number of degrees of freedom (which is not unusual under the empirical approach) allows choosing such a set of trading rules and parameters that would generate satisfactory results not only during the optimization period, but in the walk-forward analysis as well (we will examine this in detail in Chapter 5). Thus, practical use of the empirical approach exclusively is risky and hardly applicable in real trading.